sys/net/pf/pf.c

   1 /*
   2  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
   3  *
   4  * Copyright (c) 2001 Daniel Hartmeier
   5  * Copyright (c) 2002 - 2008 Henning Brauer
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  *
  12  *    - Redistributions of source code must retain the above copyright
  13  *      notice, this list of conditions and the following disclaimer.
  14  *    - Redistributions in binary form must reproduce the above
  15  *      copyright notice, this list of conditions and the following
  16  *      disclaimer in the documentation and/or other materials provided
  17  *      with the distribution.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  20  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  21  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  22  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  23  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  25  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  26  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  27  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
  29  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  30  * POSSIBILITY OF SUCH DAMAGE.
  31  *
  32  * Effort sponsored in part by the Defense Advanced Research Projects
  33  * Agency (DARPA) and Air Force Research Laboratory, Air Force
  34  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
  35  *
  36  */
  37
  38 #include "opt_inet.h"
  39 #include "opt_inet6.h"
  40
  41 #include <sys/param.h>
  42 #include <sys/systm.h>
  43 #include <sys/malloc.h>
  44 #include <sys/mbuf.h>
  45 #include <sys/filio.h>
  46 #include <sys/socket.h>
  47 #include <sys/socketvar.h>
  48 #include <sys/kernel.h>
  49 #include <sys/time.h>
  50 #include <sys/sysctl.h>
  51 #include <sys/endian.h>
  52 #include <sys/proc.h>
  53 #include <sys/kthread.h>
  54 #include <sys/spinlock.h>
  55
  56 #include <sys/md5.h>
  57
  58 #include <net/if.h>
  59 #include <net/if_types.h>
  60 #include <net/bpf.h>
  61 #include <net/netisr2.h>
  62 #include <net/route.h>
  63
  64 #include <netinet/in.h>
  65 #include <netinet/in_var.h>
  66 #include <netinet/in_systm.h>
  67 #include <netinet/ip.h>
  68 #include <netinet/ip_var.h>
  69 #include <netinet/tcp.h>
  70 #include <netinet/tcp_seq.h>
  71 #include <netinet/udp.h>
  72 #include <netinet/ip_icmp.h>
  73 #include <netinet/in_pcb.h>
  74 #include <netinet/tcp_timer.h>
  75 #include <netinet/tcp_var.h>
  76 #include <netinet/udp_var.h>
  77 #include <netinet/icmp_var.h>
  78 #include <netinet/if_ether.h>
  79
  80 #include <net/pf/pfvar.h>
  81 #include <net/pf/if_pflog.h>
  82
  83 #include <net/pf/if_pfsync.h>
  84
  85 #ifdef INET6
  86 #include <netinet/ip6.h>
  87 #include <netinet/icmp6.h>
  88 #include <netinet6/nd6.h>
  89 #include <netinet6/ip6_var.h>
  90 #include <netinet6/in6_pcb.h>
  91 #endif /* INET6 */
  92
  93 #include <sys/in_cksum.h>
  94 #include <sys/ucred.h>
  95 #include <machine/limits.h>
  96 #include <sys/msgport2.h>
  97 #include <sys/spinlock2.h>
  98 #include <net/netmsg2.h>
  99 #include <net/toeplitz2.h>
 100
 101 extern int ip_optcopy(struct ip *, struct ip *);
 102 extern int debug_pfugidhack;
 103
 104 /*
 105  * pf_token - shared lock for cpu-localized operations,
 106  *            exclusive lock otherwise.
 107  *
 108  * pf_gtoken- exclusive lock used for initialization.
 109  *
 110  * pf_spin  - only used to atomically fetch and increment stateid
 111  *            on 32-bit systems.
 112  */
 113 struct lwkt_token pf_token = LWKT_TOKEN_INITIALIZER(pf_token);
 114 struct lwkt_token pf_gtoken = LWKT_TOKEN_INITIALIZER(pf_gtoken);
 115 #if __SIZEOF_LONG__ != 8
 116 struct spinlock pf_spin = SPINLOCK_INITIALIZER(pf_spin, "pf_spin");
 117 #endif
 118
 119 #define DPFPRINTF(n, x) if (pf_status.debug >= (n)) kprintf x
 120
 121 #define FAIL(code)      { error = (code); goto done; }
 122
 123 /*
 124  * Global variables
 125  */
 126
 127 /* mask radix tree */
 128 struct radix_node_head  *pf_maskhead;
 129
 130 /* state tables */
 131 struct pf_state_tree     *pf_statetbl;          /* incls one global table */
 132 struct pf_state         **purge_cur;
 133 struct pf_altqqueue      pf_altqs[2];
 134 struct pf_palist         pf_pabuf;
 135 struct pf_altqqueue     *pf_altqs_active;
 136 struct pf_altqqueue     *pf_altqs_inactive;
 137 struct pf_status         pf_status;
 138
 139 u_int32_t                ticket_altqs_active;
 140 u_int32_t                ticket_altqs_inactive;
 141 int                      altqs_inactive_open;
 142 u_int32_t                ticket_pabuf;
 143
 144 MD5_CTX                  pf_tcp_secret_ctx;
 145 u_char                   pf_tcp_secret[16];
 146 int                      pf_tcp_secret_init;
 147 int                      pf_tcp_iss_off;
 148
 149 struct pf_anchor_stackframe {
 150         struct pf_ruleset                       *rs;
 151         struct pf_rule                          *r;
 152         struct pf_anchor_node                   *parent;
 153         struct pf_anchor                        *child;
 154 } pf_anchor_stack[64];
 155
 156 struct malloc_type       *pf_src_tree_pl, *pf_rule_pl, *pf_pooladdr_pl;
 157 struct malloc_type       *pf_state_pl, *pf_state_key_pl, *pf_state_item_pl;
 158 struct malloc_type       *pf_altq_pl;
 159
 160 void                     pf_print_host(struct pf_addr *, u_int16_t, u_int8_t);
 161
 162 void                     pf_init_threshold(struct pf_threshold *, u_int32_t,
 163                             u_int32_t);
 164 void                     pf_add_threshold(struct pf_threshold *);
 165 int                      pf_check_threshold(struct pf_threshold *);
 166
 167 void                     pf_change_ap(struct pf_addr *, u_int16_t *,
 168                             u_int16_t *, u_int16_t *, struct pf_addr *,
 169                             u_int16_t, u_int8_t, sa_family_t);
 170 int                      pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 171                             struct tcphdr *, struct pf_state_peer *);
 172 #ifdef INET6
 173 void                     pf_change_a6(struct pf_addr *, u_int16_t *,
 174                             struct pf_addr *, u_int8_t);
 175 #endif /* INET6 */
 176 void                     pf_change_icmp(struct pf_addr *, u_int16_t *,
 177                             struct pf_addr *, struct pf_addr *, u_int16_t,
 178                             u_int16_t *, u_int16_t *, u_int16_t *,
 179                             u_int16_t *, u_int8_t, sa_family_t);
 180 void                     pf_send_tcp(const struct pf_rule *, sa_family_t,
 181                             const struct pf_addr *, const struct pf_addr *,
 182                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 183                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 184                             u_int16_t, struct ether_header *, struct ifnet *);
 185 void                     pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 186                             sa_family_t, struct pf_rule *);
 187 struct pf_rule          *pf_match_translation(struct pf_pdesc *, struct mbuf *,
 188                             int, int, struct pfi_kif *,
 189                             struct pf_addr *, u_int16_t, struct pf_addr *,
 190                             u_int16_t, int);
 191 struct pf_rule          *pf_get_translation(struct pf_pdesc *, struct mbuf *,
 192                             int, int, struct pfi_kif *, struct pf_src_node **,
 193                             struct pf_state_key **, struct pf_state_key **,
 194                             struct pf_state_key **, struct pf_state_key **,
 195                             struct pf_addr *, struct pf_addr *,
 196                             u_int16_t, u_int16_t);
 197 void                     pf_detach_state(struct pf_state *);
 198 int                      pf_state_key_setup(struct pf_pdesc *, struct pf_rule *,
 199                             struct pf_state_key **, struct pf_state_key **,
 200                             struct pf_state_key **, struct pf_state_key **,
 201                             struct pf_addr *, struct pf_addr *,
 202                             u_int16_t, u_int16_t);
 203 void                     pf_state_key_detach(struct pf_state *, int);
 204 u_int32_t                pf_tcp_iss(struct pf_pdesc *);
 205 int                      pf_test_rule(struct pf_rule **, struct pf_state **,
 206                             int, struct pfi_kif *, struct mbuf *, int,
 207                             void *, struct pf_pdesc *, struct pf_rule **,
 208                             struct pf_ruleset **, struct ifqueue *, struct inpcb *);
 209 static __inline int      pf_create_state(struct pf_rule *, struct pf_rule *,
 210                             struct pf_rule *, struct pf_pdesc *,
 211                             struct pf_src_node *, struct pf_state_key *,
 212                             struct pf_state_key *, struct pf_state_key *,
 213                             struct pf_state_key *, struct mbuf *, int,
 214                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 215                             struct pf_state **, int, u_int16_t, u_int16_t,
 216                             int);
 217 int                      pf_test_fragment(struct pf_rule **, int,
 218                             struct pfi_kif *, struct mbuf *, void *,
 219                             struct pf_pdesc *, struct pf_rule **,
 220                             struct pf_ruleset **);
 221 int                      pf_tcp_track_full(struct pf_state_peer *,
 222                             struct pf_state_peer *, struct pf_state **,
 223                             struct pfi_kif *, struct mbuf *, int,
 224                             struct pf_pdesc *, u_short *, int *);
 225 int                     pf_tcp_track_sloppy(struct pf_state_peer *,
 226                             struct pf_state_peer *, struct pf_state **,
 227                             struct pf_pdesc *, u_short *);
 228 int                      pf_test_state_tcp(struct pf_state **, int,
 229                             struct pfi_kif *, struct mbuf *, int,
 230                             void *, struct pf_pdesc *, u_short *);
 231 int                      pf_test_state_udp(struct pf_state **, int,
 232                             struct pfi_kif *, struct mbuf *, int,
 233                             void *, struct pf_pdesc *);
 234 int                      pf_test_state_icmp(struct pf_state **, int,
 235                             struct pfi_kif *, struct mbuf *, int,
 236                             void *, struct pf_pdesc *, u_short *);
 237 int                      pf_test_state_other(struct pf_state **, int,
 238                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 239 void                     pf_step_into_anchor(int *, struct pf_ruleset **, int,
 240                             struct pf_rule **, struct pf_rule **, int *);
 241 int                      pf_step_out_of_anchor(int *, struct pf_ruleset **,
 242                              int, struct pf_rule **, struct pf_rule **,
 243                              int *);
 244 void                     pf_hash(struct pf_addr *, struct pf_addr *,
 245                             struct pf_poolhashkey *, sa_family_t);
 246 int                      pf_map_addr(u_int8_t, struct pf_rule *,
 247                             struct pf_addr *, struct pf_addr *,
 248                             struct pf_addr *, struct pf_src_node **);
 249 int                      pf_get_sport(struct pf_pdesc *,
 250                             sa_family_t, u_int8_t, struct pf_rule *,
 251                             struct pf_addr *, struct pf_addr *,
 252                             u_int16_t, u_int16_t,
 253                             struct pf_addr *, u_int16_t *,
 254                             u_int16_t, u_int16_t,
 255                             struct pf_src_node **);
 256 void                     pf_route(struct mbuf **, struct pf_rule *, int,
 257                             struct ifnet *, struct pf_state *,
 258                             struct pf_pdesc *);
 259 void                     pf_route6(struct mbuf **, struct pf_rule *, int,
 260                             struct ifnet *, struct pf_state *,
 261                             struct pf_pdesc *);
 262 u_int8_t                 pf_get_wscale(struct mbuf *, int, u_int16_t,
 263                             sa_family_t);
 264 u_int16_t                pf_get_mss(struct mbuf *, int, u_int16_t,
 265                             sa_family_t);
 266 u_int16_t                pf_calc_mss(struct pf_addr *, sa_family_t,
 267                                 u_int16_t);
 268 void                     pf_set_rt_ifp(struct pf_state *,
 269                             struct pf_addr *);
 270 int                      pf_check_proto_cksum(struct mbuf *, int, int,
 271                             u_int8_t, sa_family_t);
 272 struct pf_divert        *pf_get_divert(struct mbuf *);
 273 void                     pf_print_state_parts(struct pf_state *,
 274                             struct pf_state_key *, struct pf_state_key *);
 275 int                      pf_addr_wrap_neq(struct pf_addr_wrap *,
 276                             struct pf_addr_wrap *);
 277 struct pf_state         *pf_find_state(struct pfi_kif *,
 278                             struct pf_state_key_cmp *, u_int, struct mbuf *);
 279 int                      pf_src_connlimit(struct pf_state *);
 280 int                      pf_check_congestion(struct ifqueue *);
 281
 282 extern int pf_end_threads;
 283
 284 struct pf_pool_limit pf_pool_limits[PF_LIMIT_MAX] = {
 285         { &pf_state_pl, PFSTATE_HIWAT },
 286         { &pf_src_tree_pl, PFSNODE_HIWAT },
 287         { &pf_frent_pl, PFFRAG_FRENT_HIWAT },
 288         { &pfr_ktable_pl, PFR_KTABLE_HIWAT },
 289         { &pfr_kentry_pl, PFR_KENTRY_HIWAT }
 290 };
 291
 292 /*
 293  * If route-to and direction is out we match with no further processing
 294  *      (rt_kif must be assigned and not equal to the out interface)
 295  * If reply-to and direction is in we match with no further processing
 296  *      (rt_kif must be assigned and not equal to the in interface)
 297  */
 298 #define STATE_LOOKUP(i, k, d, s, m)                                     \
 299         do {                                                            \
 300                 s = pf_find_state(i, k, d, m);                          \
 301                 if (s == NULL || (s)->timeout == PFTM_PURGE)            \
 302                         return (PF_DROP);                               \
 303                 if (d == PF_OUT &&                                      \
 304                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 305                     (s)->rule.ptr->direction == PF_OUT) ||              \
 306                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 307                     (s)->rule.ptr->direction == PF_IN)) &&              \
 308                     (s)->rt_kif != NULL &&                              \
 309                     (s)->rt_kif != i)                                   \
 310                         return (PF_PASS);                               \
 311         } while (0)
 312
 313 #define BOUND_IFACE(r, k) \
 314         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : pfi_all
 315
 316 #define STATE_INC_COUNTERS(s)                           \
 317         do {                                            \
 318                 atomic_add_int(&s->rule.ptr->states_cur, 1);    \
 319                 s->rule.ptr->states_tot++;              \
 320                 if (s->anchor.ptr != NULL) {            \
 321                         atomic_add_int(&s->anchor.ptr->states_cur, 1);  \
 322                         s->anchor.ptr->states_tot++;    \
 323                 }                                       \
 324                 if (s->nat_rule.ptr != NULL) {          \
 325                         atomic_add_int(&s->nat_rule.ptr->states_cur, 1); \
 326                         s->nat_rule.ptr->states_tot++;  \
 327                 }                                       \
 328         } while (0)
 329
 330 #define STATE_DEC_COUNTERS(s)                           \
 331         do {                                            \
 332                 if (s->nat_rule.ptr != NULL)            \
 333                         atomic_add_int(&s->nat_rule.ptr->states_cur, -1); \
 334                 if (s->anchor.ptr != NULL)              \
 335                         atomic_add_int(&s->anchor.ptr->states_cur, -1); \
 336                 atomic_add_int(&s->rule.ptr->states_cur, -1);           \
 337         } while (0)
 338
 339 static MALLOC_DEFINE(M_PFSTATEPL, "pfstatepl", "pf state pool list");
 340 static MALLOC_DEFINE(M_PFSRCTREEPL, "pfsrctpl", "pf source tree pool list");
 341 static MALLOC_DEFINE(M_PFSTATEKEYPL, "pfstatekeypl", "pf state key pool list");
 342 static MALLOC_DEFINE(M_PFSTATEITEMPL, "pfstateitempl", "pf state item pool list");
 343
 344 static __inline int pf_src_compare(struct pf_src_node *, struct pf_src_node *);
 345 static __inline int pf_state_compare_key(struct pf_state_key *,
 346                                 struct pf_state_key *);
 347 static __inline int pf_state_compare_rkey(struct pf_state_key *,
 348                                 struct pf_state_key *);
 349 static __inline int pf_state_compare_id(struct pf_state *,
 350                                 struct pf_state *);
 351
 352 struct pf_src_tree *tree_src_tracking;
 353 struct pf_state_tree_id *tree_id;
 354 struct pf_state_queue *state_list;
 355 struct pf_counters *pf_counters;
 356
 357 RB_GENERATE(pf_src_tree, pf_src_node, entry, pf_src_compare);
 358 RB_GENERATE(pf_state_tree, pf_state_key, entry, pf_state_compare_key);
 359 RB_GENERATE(pf_state_rtree, pf_state_key, entry, pf_state_compare_rkey);
 360 RB_GENERATE(pf_state_tree_id, pf_state, entry_id, pf_state_compare_id);
 361
 362 static __inline int
 363 pf_src_compare(struct pf_src_node *a, struct pf_src_node *b)
 364 {
 365         int     diff;
 366
 367         if (a->rule.ptr > b->rule.ptr)
 368                 return (1);
 369         if (a->rule.ptr < b->rule.ptr)
 370                 return (-1);
 371         if ((diff = a->af - b->af) != 0)
 372                 return (diff);
 373         switch (a->af) {
 374 #ifdef INET
 375         case AF_INET:
 376                 if (a->addr.addr32[0] > b->addr.addr32[0])
 377                         return (1);
 378                 if (a->addr.addr32[0] < b->addr.addr32[0])
 379                         return (-1);
 380                 break;
 381 #endif /* INET */
 382 #ifdef INET6
 383         case AF_INET6:
 384                 if (a->addr.addr32[3] > b->addr.addr32[3])
 385                         return (1);
 386                 if (a->addr.addr32[3] < b->addr.addr32[3])
 387                         return (-1);
 388                 if (a->addr.addr32[2] > b->addr.addr32[2])
 389                         return (1);
 390                 if (a->addr.addr32[2] < b->addr.addr32[2])
 391                         return (-1);
 392                 if (a->addr.addr32[1] > b->addr.addr32[1])
 393                         return (1);
 394                 if (a->addr.addr32[1] < b->addr.addr32[1])
 395                         return (-1);
 396                 if (a->addr.addr32[0] > b->addr.addr32[0])
 397                         return (1);
 398                 if (a->addr.addr32[0] < b->addr.addr32[0])
 399                         return (-1);
 400                 break;
 401 #endif /* INET6 */
 402         }
 403         return (0);
 404 }
 405
 406 u_int32_t
 407 pf_state_hash(struct pf_state_key *sk)
 408 {
 409         u_int32_t hv = (u_int32_t)(((intptr_t)sk >> 6) ^ ((intptr_t)sk >> 15));
 410         if (hv == 0)    /* disallow 0 */
 411                 hv = 1;
 412         return(hv);
 413 }
 414
 415 #ifdef INET6
 416 void
 417 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 418 {
 419         switch (af) {
 420 #ifdef INET
 421         case AF_INET:
 422                 dst->addr32[0] = src->addr32[0];
 423                 break;
 424 #endif /* INET */
 425         case AF_INET6:
 426                 dst->addr32[0] = src->addr32[0];
 427                 dst->addr32[1] = src->addr32[1];
 428                 dst->addr32[2] = src->addr32[2];
 429                 dst->addr32[3] = src->addr32[3];
 430                 break;
 431         }
 432 }
 433 #endif /* INET6 */
 434
 435 void
 436 pf_init_threshold(struct pf_threshold *threshold,
 437     u_int32_t limit, u_int32_t seconds)
 438 {
 439         threshold->limit = limit * PF_THRESHOLD_MULT;
 440         threshold->seconds = seconds;
 441         threshold->count = 0;
 442         threshold->last = time_second;
 443 }
 444
 445 void
 446 pf_add_threshold(struct pf_threshold *threshold)
 447 {
 448         u_int32_t t = time_second, diff = t - threshold->last;
 449
 450         if (diff >= threshold->seconds)
 451                 threshold->count = 0;
 452         else
 453                 threshold->count -= threshold->count * diff /
 454                     threshold->seconds;
 455         threshold->count += PF_THRESHOLD_MULT;
 456         threshold->last = t;
 457 }
 458
 459 int
 460 pf_check_threshold(struct pf_threshold *threshold)
 461 {
 462         return (threshold->count > threshold->limit);
 463 }
 464
 465 int
 466 pf_src_connlimit(struct pf_state *state)
 467 {
 468         int bad = 0;
 469         int cpu = mycpu->gd_cpuid;
 470
 471         atomic_add_int(&state->src_node->conn, 1);
 472         state->src.tcp_est = 1;
 473         pf_add_threshold(&state->src_node->conn_rate);
 474
 475         if (state->rule.ptr->max_src_conn &&
 476             state->rule.ptr->max_src_conn <
 477             state->src_node->conn) {
 478                 PF_INC_LCOUNTER(LCNT_SRCCONN);
 479                 bad++;
 480         }
 481
 482         if (state->rule.ptr->max_src_conn_rate.limit &&
 483             pf_check_threshold(&state->src_node->conn_rate)) {
 484                 PF_INC_LCOUNTER(LCNT_SRCCONNRATE);
 485                 bad++;
 486         }
 487
 488         if (!bad)
 489                 return 0;
 490
 491         if (state->rule.ptr->overload_tbl) {
 492                 struct pfr_addr p;
 493                 u_int32_t       killed = 0;
 494
 495                 PF_INC_LCOUNTER(LCNT_OVERLOAD_TABLE);
 496                 if (pf_status.debug >= PF_DEBUG_MISC) {
 497                         kprintf("pf_src_connlimit: blocking address ");
 498                         pf_print_host(&state->src_node->addr, 0,
 499                             state->key[PF_SK_WIRE]->af);
 500                 }
 501
 502                 bzero(&p, sizeof(p));
 503                 p.pfra_af = state->key[PF_SK_WIRE]->af;
 504                 switch (state->key[PF_SK_WIRE]->af) {
 505 #ifdef INET
 506                 case AF_INET:
 507                         p.pfra_net = 32;
 508                         p.pfra_ip4addr = state->src_node->addr.v4;
 509                         break;
 510 #endif /* INET */
 511 #ifdef INET6
 512                 case AF_INET6:
 513                         p.pfra_net = 128;
 514                         p.pfra_ip6addr = state->src_node->addr.v6;
 515                         break;
 516 #endif /* INET6 */
 517                 }
 518
 519                 pfr_insert_kentry(state->rule.ptr->overload_tbl,
 520                     &p, time_second);
 521
 522                 /* kill existing states if that's required. */
 523                 if (state->rule.ptr->flush) {
 524                         struct pf_state_key *sk;
 525                         struct pf_state *st;
 526
 527                         PF_INC_LCOUNTER(LCNT_OVERLOAD_FLUSH);
 528                         RB_FOREACH(st, pf_state_tree_id, &tree_id[cpu]) {
 529                                 sk = st->key[PF_SK_WIRE];
 530                                 /*
 531                                  * Kill states from this source.  (Only those
 532                                  * from the same rule if PF_FLUSH_GLOBAL is not
 533                                  * set).  (Only on current cpu).
 534                                  */
 535                                 if (sk->af ==
 536                                     state->key[PF_SK_WIRE]->af &&
 537                                     ((state->direction == PF_OUT &&
 538                                     PF_AEQ(&state->src_node->addr,
 539                                         &sk->addr[0], sk->af)) ||
 540                                     (state->direction == PF_IN &&
 541                                     PF_AEQ(&state->src_node->addr,
 542                                         &sk->addr[1], sk->af))) &&
 543                                     (state->rule.ptr->flush &
 544                                     PF_FLUSH_GLOBAL ||
 545                                     state->rule.ptr == st->rule.ptr)) {
 546                                         st->timeout = PFTM_PURGE;
 547                                         st->src.state = st->dst.state =
 548                                             TCPS_CLOSED;
 549                                         killed++;
 550                                 }
 551                         }
 552                         if (pf_status.debug >= PF_DEBUG_MISC)
 553                                 kprintf(", %u states killed", killed);
 554                 }
 555                 if (pf_status.debug >= PF_DEBUG_MISC)
 556                         kprintf("\n");
 557         }
 558
 559         /* kill this state */
 560         state->timeout = PFTM_PURGE;
 561         state->src.state = state->dst.state = TCPS_CLOSED;
 562
 563         return 1;
 564 }
 565
 566 int
 567 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 568     struct pf_addr *src, sa_family_t af)
 569 {
 570         struct pf_src_node      k;
 571         int cpu = mycpu->gd_cpuid;
 572
 573         bzero(&k, sizeof(k));   /* avoid gcc warnings */
 574         if (*sn == NULL) {
 575                 k.af = af;
 576                 PF_ACPY(&k.addr, src, af);
 577                 if (rule->rule_flag & PFRULE_RULESRCTRACK ||
 578                     rule->rpool.opts & PF_POOL_STICKYADDR)
 579                         k.rule.ptr = rule;
 580                 else
 581                         k.rule.ptr = NULL;
 582                 PF_INC_SCOUNTER(SCNT_SRC_NODE_SEARCH);
 583                 *sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k);
 584         }
 585         if (*sn == NULL) {
 586                 if (!rule->max_src_nodes ||
 587                     rule->src_nodes < rule->max_src_nodes)
 588                         (*sn) = kmalloc(sizeof(struct pf_src_node),
 589                                         M_PFSRCTREEPL, M_NOWAIT|M_ZERO);
 590                 else
 591                         PF_INC_LCOUNTER(LCNT_SRCNODES);
 592                 if ((*sn) == NULL)
 593                         return (-1);
 594
 595                 pf_init_threshold(&(*sn)->conn_rate,
 596                     rule->max_src_conn_rate.limit,
 597                     rule->max_src_conn_rate.seconds);
 598
 599                 (*sn)->af = af;
 600                 if (rule->rule_flag & PFRULE_RULESRCTRACK ||
 601                     rule->rpool.opts & PF_POOL_STICKYADDR)
 602                         (*sn)->rule.ptr = rule;
 603                 else
 604                         (*sn)->rule.ptr = NULL;
 605                 PF_ACPY(&(*sn)->addr, src, af);
 606                 if (RB_INSERT(pf_src_tree,
 607                     &tree_src_tracking[cpu], *sn) != NULL) {
 608                         if (pf_status.debug >= PF_DEBUG_MISC) {
 609                                 kprintf("pf: src_tree insert failed: ");
 610                                 pf_print_host(&(*sn)->addr, 0, af);
 611                                 kprintf("\n");
 612                         }
 613                         kfree(*sn, M_PFSRCTREEPL);
 614                         return (-1);
 615                 }
 616
 617                 /*
 618                  * Atomic op required to increment src_nodes in the rule
 619                  * because we hold a shared token here (decrements will use
 620                  * an exclusive token).
 621                  */
 622                 (*sn)->creation = time_second;
 623                 (*sn)->ruletype = rule->action;
 624                 if ((*sn)->rule.ptr != NULL)
 625                         atomic_add_int(&(*sn)->rule.ptr->src_nodes, 1);
 626                 PF_INC_SCOUNTER(SCNT_SRC_NODE_INSERT);
 627                 atomic_add_int(&pf_status.src_nodes, 1);
 628         } else {
 629                 if (rule->max_src_states &&
 630                     (*sn)->states >= rule->max_src_states) {
 631                         PF_INC_LCOUNTER(LCNT_SRCSTATES);
 632                         return (-1);
 633                 }
 634         }
 635         return (0);
 636 }
 637
 638 /*
 639  * state table (indexed by the pf_state_key structure), normal RBTREE
 640  * comparison.
 641  */
 642 static __inline int
 643 pf_state_compare_key(struct pf_state_key *a, struct pf_state_key *b)
 644 {
 645         int     diff;
 646
 647         if ((diff = a->proto - b->proto) != 0)
 648                 return (diff);
 649         if ((diff = a->af - b->af) != 0)
 650                 return (diff);
 651         switch (a->af) {
 652 #ifdef INET
 653         case AF_INET:
 654                 if (a->addr[0].addr32[0] > b->addr[0].addr32[0])
 655                         return (1);
 656                 if (a->addr[0].addr32[0] < b->addr[0].addr32[0])
 657                         return (-1);
 658                 if (a->addr[1].addr32[0] > b->addr[1].addr32[0])
 659                         return (1);
 660                 if (a->addr[1].addr32[0] < b->addr[1].addr32[0])
 661                         return (-1);
 662                 break;
 663 #endif /* INET */
 664 #ifdef INET6
 665         case AF_INET6:
 666                 if (a->addr[0].addr32[3] > b->addr[0].addr32[3])
 667                         return (1);
 668                 if (a->addr[0].addr32[3] < b->addr[0].addr32[3])
 669                         return (-1);
 670                 if (a->addr[1].addr32[3] > b->addr[1].addr32[3])
 671                         return (1);
 672                 if (a->addr[1].addr32[3] < b->addr[1].addr32[3])
 673                         return (-1);
 674                 if (a->addr[0].addr32[2] > b->addr[0].addr32[2])
 675                         return (1);
 676                 if (a->addr[0].addr32[2] < b->addr[0].addr32[2])
 677                         return (-1);
 678                 if (a->addr[1].addr32[2] > b->addr[1].addr32[2])
 679                         return (1);
 680                 if (a->addr[1].addr32[2] < b->addr[1].addr32[2])
 681                         return (-1);
 682                 if (a->addr[0].addr32[1] > b->addr[0].addr32[1])
 683                         return (1);
 684                 if (a->addr[0].addr32[1] < b->addr[0].addr32[1])
 685                         return (-1);
 686                 if (a->addr[1].addr32[1] > b->addr[1].addr32[1])
 687                         return (1);
 688                 if (a->addr[1].addr32[1] < b->addr[1].addr32[1])
 689                         return (-1);
 690                 if (a->addr[0].addr32[0] > b->addr[0].addr32[0])
 691                         return (1);
 692                 if (a->addr[0].addr32[0] < b->addr[0].addr32[0])
 693                         return (-1);
 694                 if (a->addr[1].addr32[0] > b->addr[1].addr32[0])
 695                         return (1);
 696                 if (a->addr[1].addr32[0] < b->addr[1].addr32[0])
 697                         return (-1);
 698                 break;
 699 #endif /* INET6 */
 700         }
 701
 702         if ((diff = a->port[0] - b->port[0]) != 0)
 703                 return (diff);
 704         if ((diff = a->port[1] - b->port[1]) != 0)
 705                 return (diff);
 706
 707         return (0);
 708 }
 709
 710 /*
 711  * Used for RB_FIND only, compare in the reverse direction.  The
 712  * element to be reversed is always (a), since we obviously can't
 713  * reverse the state tree depicted by (b).
 714  */
 715 static __inline int
 716 pf_state_compare_rkey(struct pf_state_key *a, struct pf_state_key *b)
 717 {
 718         int     diff;
 719
 720         if ((diff = a->proto - b->proto) != 0)
 721                 return (diff);
 722         if ((diff = a->af - b->af) != 0)
 723                 return (diff);
 724         switch (a->af) {
 725 #ifdef INET
 726         case AF_INET:
 727                 if (a->addr[1].addr32[0] > b->addr[0].addr32[0])
 728                         return (1);
 729                 if (a->addr[1].addr32[0] < b->addr[0].addr32[0])
 730                         return (-1);
 731                 if (a->addr[0].addr32[0] > b->addr[1].addr32[0])
 732                         return (1);
 733                 if (a->addr[0].addr32[0] < b->addr[1].addr32[0])
 734                         return (-1);
 735                 break;
 736 #endif /* INET */
 737 #ifdef INET6
 738         case AF_INET6:
 739                 if (a->addr[1].addr32[3] > b->addr[0].addr32[3])
 740                         return (1);
 741                 if (a->addr[1].addr32[3] < b->addr[0].addr32[3])
 742                         return (-1);
 743                 if (a->addr[0].addr32[3] > b->addr[1].addr32[3])
 744                         return (1);
 745                 if (a->addr[0].addr32[3] < b->addr[1].addr32[3])
 746                         return (-1);
 747                 if (a->addr[1].addr32[2] > b->addr[0].addr32[2])
 748                         return (1);
 749                 if (a->addr[1].addr32[2] < b->addr[0].addr32[2])
 750                         return (-1);
 751                 if (a->addr[0].addr32[2] > b->addr[1].addr32[2])
 752                         return (1);
 753                 if (a->addr[0].addr32[2] < b->addr[1].addr32[2])
 754                         return (-1);
 755                 if (a->addr[1].addr32[1] > b->addr[0].addr32[1])
 756                         return (1);
 757                 if (a->addr[1].addr32[1] < b->addr[0].addr32[1])
 758                         return (-1);
 759                 if (a->addr[0].addr32[1] > b->addr[1].addr32[1])
 760                         return (1);
 761                 if (a->addr[0].addr32[1] < b->addr[1].addr32[1])
 762                         return (-1);
 763                 if (a->addr[1].addr32[0] > b->addr[0].addr32[0])
 764                         return (1);
 765                 if (a->addr[1].addr32[0] < b->addr[0].addr32[0])
 766                         return (-1);
 767                 if (a->addr[0].addr32[0] > b->addr[1].addr32[0])
 768                         return (1);
 769                 if (a->addr[0].addr32[0] < b->addr[1].addr32[0])
 770                         return (-1);
 771                 break;
 772 #endif /* INET6 */
 773         }
 774
 775         if ((diff = a->port[1] - b->port[0]) != 0)
 776                 return (diff);
 777         if ((diff = a->port[0] - b->port[1]) != 0)
 778                 return (diff);
 779
 780         return (0);
 781 }
 782
 783 static __inline int
 784 pf_state_compare_id(struct pf_state *a, struct pf_state *b)
 785 {
 786         if (a->id > b->id)
 787                 return (1);
 788         if (a->id < b->id)
 789                 return (-1);
 790         if (a->creatorid > b->creatorid)
 791                 return (1);
 792         if (a->creatorid < b->creatorid)
 793                 return (-1);
 794
 795         return (0);
 796 }
 797
 798 int
 799 pf_state_key_attach(struct pf_state_key *sk, struct pf_state *s, int idx)
 800 {
 801         struct pf_state_item    *si;
 802         struct pf_state_key     *cur;
 803         int cpu;
 804         int error;
 805
 806         /*
 807          * PFSTATE_STACK_GLOBAL is set when the state might not hash to the
 808          * current cpu.  The keys are managed on the global statetbl tree
 809          * for this case.  Only translations (RDR, NAT) can cause this.
 810          *
 811          * When this flag is not set we must still check the global statetbl
 812          * for a collision, and if we find one we set the HALF_DUPLEX flag
 813          * in the state.
 814          */
 815         if (s->state_flags & PFSTATE_STACK_GLOBAL) {
 816                 cpu = ncpus;
 817                 lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE);
 818         } else {
 819                 cpu = mycpu->gd_cpuid;
 820                 lockmgr(&pf_global_statetbl_lock, LK_SHARED);
 821         }
 822         KKASSERT(s->key[idx] == NULL);  /* XXX handle this? */
 823
 824         if (pf_status.debug >= PF_DEBUG_MISC) {
 825                 kprintf("state_key attach cpu %d (%08x:%d) %s (%08x:%d)\n",
 826                         cpu,
 827                         ntohl(sk->addr[0].addr32[0]), ntohs(sk->port[0]),
 828                         (idx == PF_SK_WIRE ? "->" : "<-"),
 829                         ntohl(sk->addr[1].addr32[0]), ntohs(sk->port[1]));
 830         }
 831
 832         /*
 833          * Check whether (e.g.) a PASS rule being put on a per-cpu tree
 834          * collides with a translation rule on the global tree.  This is
 835          * NOT an error.  We *WANT* to establish state for this case so the
 836          * packet path is short-cutted and doesn't need to scan the ruleset
 837          * on every packet.  But the established state will only see one
 838          * side of a two-way packet conversation.  To prevent this from
 839          * causing problems (e.g. generating a RST), we force PFSTATE_SLOPPY
 840          * to be set on the established state.
 841          *
 842          * A collision against RDR state can only occur with a PASS IN in the
 843          * opposite direction or a PASS OUT in the forwards direction.  This
 844          * is because RDRs are processed on the input side.
 845          *
 846          * A collision against NAT state can only occur with a PASS IN in the
 847          * forwards direction or a PASS OUT in the opposite direction.  This
 848          * is because NATs are processed on the output side.
 849          *
 850          * In both situations we need to do a reverse addr/port test because
 851          * the PASS IN or PASS OUT only establishes if it doesn't match the
 852          * established RDR state in the forwards direction.  The direction
 853          * flag has to be ignored (it will be one way for a PASS IN and the
 854          * other way for a PASS OUT).
 855          *
 856          * pf_global_statetbl_lock will be locked shared when testing and
 857          * not entering into the global state table.
 858          */
 859         if (cpu != ncpus &&
 860             (cur = RB_FIND(pf_state_rtree,
 861                            (struct pf_state_rtree *)&pf_statetbl[ncpus],
 862                            sk)) != NULL) {
 863                 TAILQ_FOREACH(si, &cur->states, entry) {
 864                         /*
 865                          * NOTE: We must ignore direction mismatches.
 866                          */
 867                         if (si->s->kif == s->kif) {
 868                                 s->state_flags |= PFSTATE_HALF_DUPLEX |
 869                                                   PFSTATE_SLOPPY;
 870                                 if (pf_status.debug >= PF_DEBUG_MISC) {
 871                                         kprintf(
 872                                             "pf: %s key attach collision "
 873                                             "on %s: ",
 874                                             (idx == PF_SK_WIRE) ?
 875                                             "wire" : "stack",
 876                                             s->kif->pfik_name);
 877                                         pf_print_state_parts(s,
 878                                             (idx == PF_SK_WIRE) ? sk : NULL,
 879                                             (idx == PF_SK_STACK) ? sk : NULL);
 880                                         kprintf("\n");
 881                                 }
 882                                 break;
 883                         }
 884                 }
 885         }
 886
 887         /*
 888          * Enter into either the per-cpu or the global state table.
 889          *
 890          * pf_global_statetbl_lock will be locked exclusively when entering
 891          * into the global state table.
 892          */
 893         if ((cur = RB_INSERT(pf_state_tree, &pf_statetbl[cpu], sk)) != NULL) {
 894                 /* key exists. check for same kif, if none, add to key */
 895                 TAILQ_FOREACH(si, &cur->states, entry) {
 896                         if (si->s->kif == s->kif &&
 897                             si->s->direction == s->direction) {
 898                                 if (pf_status.debug >= PF_DEBUG_MISC) {
 899                                         kprintf(
 900                                             "pf: %s key attach failed on %s: ",
 901                                             (idx == PF_SK_WIRE) ?
 902                                             "wire" : "stack",
 903                                             s->kif->pfik_name);
 904                                         pf_print_state_parts(s,
 905                                             (idx == PF_SK_WIRE) ? sk : NULL,
 906                                             (idx == PF_SK_STACK) ? sk : NULL);
 907                                         kprintf("\n");
 908                                 }
 909                                 kfree(sk, M_PFSTATEKEYPL);
 910                                 error = -1;
 911                                 goto failed;    /* collision! */
 912                         }
 913                 }
 914                 kfree(sk, M_PFSTATEKEYPL);
 915
 916                 s->key[idx] = cur;
 917         } else {
 918                 s->key[idx] = sk;
 919         }
 920
 921         if ((si = kmalloc(sizeof(struct pf_state_item),
 922                           M_PFSTATEITEMPL, M_NOWAIT)) == NULL) {
 923                 pf_state_key_detach(s, idx);
 924                 error = -1;
 925                 goto failed;    /* collision! */
 926         }
 927         si->s = s;
 928
 929         /* list is sorted, if-bound states before floating */
 930         if (s->kif == pfi_all)
 931                 TAILQ_INSERT_TAIL(&s->key[idx]->states, si, entry);
 932         else
 933                 TAILQ_INSERT_HEAD(&s->key[idx]->states, si, entry);
 934
 935         error = 0;
 936 failed:
 937         lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
 938         return error;
 939 }
 940
 941 /*
 942  * NOTE: Can only be called indirectly via the purge thread with pf_token
 943  *       exclusively locked.
 944  */
 945 void
 946 pf_detach_state(struct pf_state *s)
 947 {
 948         if (s->key[PF_SK_WIRE] == s->key[PF_SK_STACK])
 949                 s->key[PF_SK_WIRE] = NULL;
 950
 951         if (s->key[PF_SK_STACK] != NULL)
 952                 pf_state_key_detach(s, PF_SK_STACK);
 953
 954         if (s->key[PF_SK_WIRE] != NULL)
 955                 pf_state_key_detach(s, PF_SK_WIRE);
 956 }
 957
 958 /*
 959  * NOTE: Can only be called indirectly via the purge thread with pf_token
 960  *       exclusively locked.
 961  */
 962 void
 963 pf_state_key_detach(struct pf_state *s, int idx)
 964 {
 965         struct pf_state_item    *si;
 966         int cpu;
 967
 968         /*
 969          * PFSTATE_STACK_GLOBAL is set for translations when the translated
 970          * address/port is not localized to the same cpu that the untranslated
 971          * address/port is on.  The wire pf_state_key is managed on the global
 972          * statetbl tree for this case.
 973          */
 974         if (s->state_flags & PFSTATE_STACK_GLOBAL) {
 975                 cpu = ncpus;
 976                 lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE);
 977         } else {
 978                 cpu = mycpu->gd_cpuid;
 979         }
 980
 981         si = TAILQ_FIRST(&s->key[idx]->states);
 982         while (si && si->s != s)
 983                 si = TAILQ_NEXT(si, entry);
 984
 985         if (si) {
 986                 TAILQ_REMOVE(&s->key[idx]->states, si, entry);
 987                 kfree(si, M_PFSTATEITEMPL);
 988         }
 989
 990         if (TAILQ_EMPTY(&s->key[idx]->states)) {
 991                 RB_REMOVE(pf_state_tree, &pf_statetbl[cpu], s->key[idx]);
 992                 if (s->key[idx]->reverse)
 993                         s->key[idx]->reverse->reverse = NULL;
 994                 if (s->key[idx]->inp)
 995                         s->key[idx]->inp->inp_pf_sk = NULL;
 996                 kfree(s->key[idx], M_PFSTATEKEYPL);
 997         }
 998         s->key[idx] = NULL;
 999
1000         if (s->state_flags & PFSTATE_STACK_GLOBAL)
1001                 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1002 }
1003
1004 struct pf_state_key *
1005 pf_alloc_state_key(int pool_flags)
1006 {
1007         struct pf_state_key     *sk;
1008
1009         sk = kmalloc(sizeof(struct pf_state_key), M_PFSTATEKEYPL, pool_flags);
1010         if (sk) {
1011                 TAILQ_INIT(&sk->states);
1012         }
1013         return (sk);
1014 }
1015
1016 int
1017 pf_state_key_setup(struct pf_pdesc *pd, struct pf_rule *nr,
1018         struct pf_state_key **skw, struct pf_state_key **sks,
1019         struct pf_state_key **skp, struct pf_state_key **nkp,
1020         struct pf_addr *saddr, struct pf_addr *daddr,
1021         u_int16_t sport, u_int16_t dport)
1022 {
1023         KKASSERT((*skp == NULL && *nkp == NULL));
1024
1025         if ((*skp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL)
1026                 return (ENOMEM);
1027
1028         PF_ACPY(&(*skp)->addr[pd->sidx], saddr, pd->af);
1029         PF_ACPY(&(*skp)->addr[pd->didx], daddr, pd->af);
1030         (*skp)->port[pd->sidx] = sport;
1031         (*skp)->port[pd->didx] = dport;
1032         (*skp)->proto = pd->proto;
1033         (*skp)->af = pd->af;
1034
1035         if (nr != NULL) {
1036                 if ((*nkp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL)
1037                         return (ENOMEM); /* caller must handle cleanup */
1038
1039                 /* XXX maybe just bcopy and TAILQ_INIT(&(*nkp)->states) */
1040                 PF_ACPY(&(*nkp)->addr[0], &(*skp)->addr[0], pd->af);
1041                 PF_ACPY(&(*nkp)->addr[1], &(*skp)->addr[1], pd->af);
1042                 (*nkp)->port[0] = (*skp)->port[0];
1043                 (*nkp)->port[1] = (*skp)->port[1];
1044                 (*nkp)->proto = pd->proto;
1045                 (*nkp)->af = pd->af;
1046         } else {
1047                 *nkp = *skp;
1048         }
1049
1050         if (pd->dir == PF_IN) {
1051                 *skw = *skp;
1052                 *sks = *nkp;
1053         } else {
1054                 *sks = *skp;
1055                 *skw = *nkp;
1056         }
1057         return (0);
1058 }
1059
1060 /*
1061  * Insert pf_state with one or two state keys (allowing a reverse path lookup
1062  * which is used by NAT).  In the NAT case skw is the initiator (?) and
1063  * sks is the target.
1064  */
1065 int
1066 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1067                 struct pf_state_key *sks, struct pf_state *s)
1068 {
1069         int cpu = mycpu->gd_cpuid;
1070
1071         s->kif = kif;
1072         s->cpuid = cpu;
1073
1074         if (skw == sks) {
1075                 if (pf_state_key_attach(skw, s, PF_SK_WIRE))
1076                         return (-1);
1077                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1078         } else {
1079                 /*
1080                 skw->reverse = sks;
1081                 sks->reverse = skw;
1082                 */
1083                 if (pf_state_key_attach(skw, s, PF_SK_WIRE)) {
1084                         kfree(sks, M_PFSTATEKEYPL);
1085                         return (-1);
1086                 }
1087                 if (pf_state_key_attach(sks, s, PF_SK_STACK)) {
1088                         pf_state_key_detach(s, PF_SK_WIRE);
1089                         return (-1);
1090                 }
1091         }
1092
1093         if (s->id == 0 && s->creatorid == 0) {
1094                 u_int64_t sid;
1095
1096 #if __SIZEOF_LONG__ == 8
1097                 sid = atomic_fetchadd_long(&pf_status.stateid, 1);
1098 #else
1099                 spin_lock(&pf_spin);
1100                 sid = pf_status.stateid++;
1101                 spin_unlock(&pf_spin);
1102 #endif
1103                 s->id = htobe64(sid);
1104                 s->creatorid = pf_status.hostid;
1105         }
1106
1107         /*
1108          * Calculate hash code for altq
1109          */
1110         s->hash = crc32(s->key[PF_SK_WIRE], PF_STATE_KEY_HASH_LENGTH);
1111
1112         if (RB_INSERT(pf_state_tree_id, &tree_id[cpu], s) != NULL) {
1113                 if (pf_status.debug >= PF_DEBUG_MISC) {
1114                         kprintf("pf: state insert failed: "
1115                             "id: %016jx creatorid: %08x",
1116                               (uintmax_t)be64toh(s->id), ntohl(s->creatorid));
1117                         if (s->sync_flags & PFSTATE_FROMSYNC)
1118                                 kprintf(" (from sync)");
1119                         kprintf("\n");
1120                 }
1121                 pf_detach_state(s);
1122                 return (-1);
1123         }
1124         TAILQ_INSERT_TAIL(&state_list[cpu], s, entry_list);
1125         PF_INC_FCOUNTER(FCNT_STATE_INSERT);
1126         atomic_add_int(&pf_status.states, 1);
1127         pfi_kif_ref(kif, PFI_KIF_REF_STATE);
1128         pfsync_insert_state(s);
1129         return (0);
1130 }
1131
1132 struct pf_state *
1133 pf_find_state_byid(struct pf_state_cmp *key)
1134 {
1135         int cpu = mycpu->gd_cpuid;
1136
1137         PF_INC_FCOUNTER(FCNT_STATE_SEARCH);
1138
1139         return (RB_FIND(pf_state_tree_id, &tree_id[cpu],
1140                         (struct pf_state *)key));
1141 }
1142
1143 /*
1144  * WARNING! May return a state structure that was localized to another cpu,
1145  *          destruction is typically protected by the callers pf_token.
1146  *          The element can only be destroyed
1147  */
1148 struct pf_state *
1149 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir,
1150               struct mbuf *m)
1151 {
1152         struct pf_state_key     *skey = (void *)key;
1153         struct pf_state_key     *sk;
1154         struct pf_state_item    *si;
1155         struct pf_state *s;
1156         int cpu = mycpu->gd_cpuid;
1157         int globalstl = 0;
1158
1159         PF_INC_FCOUNTER(FCNT_STATE_SEARCH);
1160
1161         if (dir == PF_OUT && m->m_pkthdr.pf.statekey &&
1162             ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->reverse) {
1163                 sk = ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->reverse;
1164         } else {
1165                 sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey);
1166                 if (sk == NULL) {
1167                         lockmgr(&pf_global_statetbl_lock, LK_SHARED);
1168                         sk = RB_FIND(pf_state_tree, &pf_statetbl[ncpus], skey);
1169                         if (sk == NULL) {
1170                                 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1171                                 return (NULL);
1172                         }
1173                         globalstl = 1;
1174                 }
1175                 if (dir == PF_OUT && m->m_pkthdr.pf.statekey) {
1176                         ((struct pf_state_key *)
1177                             m->m_pkthdr.pf.statekey)->reverse = sk;
1178                         sk->reverse = m->m_pkthdr.pf.statekey;
1179                 }
1180         }
1181         if (dir == PF_OUT)
1182                 m->m_pkthdr.pf.statekey = NULL;
1183
1184         /* list is sorted, if-bound states before floating ones */
1185         TAILQ_FOREACH(si, &sk->states, entry) {
1186                 if ((si->s->kif == pfi_all || si->s->kif == kif) &&
1187                     sk == (dir == PF_IN ? si->s->key[PF_SK_WIRE] :
1188                                           si->s->key[PF_SK_STACK])) {
1189                         break;
1190                 }
1191         }
1192
1193         /*
1194          * Extract state before potentially releasing the global statetbl
1195          * lock.  Ignore the state if the create is still in-progress as
1196          * it can be deleted out from under us by the owning localized cpu.
1197          * However, if CREATEINPROG is not set, state can only be deleted
1198          * by the purge thread which we are protected from via our shared
1199          * pf_token.
1200          */
1201         if (si) {
1202                 s = si->s;
1203                 if (s && (s->state_flags & PFSTATE_CREATEINPROG))
1204                         s = NULL;
1205         } else {
1206                 s = NULL;
1207         }
1208         if (globalstl)
1209                 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1210         return s;
1211 }
1212
1213 /*
1214  * WARNING! May return a state structure that was localized to another cpu,
1215  *          destruction is typically protected by the callers pf_token.
1216  */
1217 struct pf_state *
1218 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1219 {
1220         struct pf_state_key     *skey = (void *)key;
1221         struct pf_state_key     *sk;
1222         struct pf_state_item    *si, *ret = NULL;
1223         struct pf_state         *s;
1224         int cpu = mycpu->gd_cpuid;
1225         int globalstl = 0;
1226
1227         PF_INC_FCOUNTER(FCNT_STATE_SEARCH);
1228
1229         sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey);
1230         if (sk == NULL) {
1231                 lockmgr(&pf_global_statetbl_lock, LK_SHARED);
1232                 sk = RB_FIND(pf_state_tree, &pf_statetbl[ncpus], skey);
1233                 globalstl = 1;
1234         }
1235         if (sk != NULL) {
1236                 TAILQ_FOREACH(si, &sk->states, entry)
1237                         if (dir == PF_INOUT ||
1238                             (sk == (dir == PF_IN ? si->s->key[PF_SK_WIRE] :
1239                             si->s->key[PF_SK_STACK]))) {
1240                                 if (more == NULL) {
1241                                         ret = si;
1242                                         break;
1243                                 }
1244                                 if (ret)
1245                                         (*more)++;
1246                                 else
1247                                         ret = si;
1248                         }
1249         }
1250
1251         /*
1252          * Extract state before potentially releasing the global statetbl
1253          * lock.  Ignore the state if the create is still in-progress as
1254          * it can be deleted out from under us by the owning localized cpu.
1255          * However, if CREATEINPROG is not set, state can only be deleted
1256          * by the purge thread which we are protected from via our shared
1257          * pf_token.
1258          */
1259         if (ret) {
1260                 s = ret->s;
1261                 if (s && (s->state_flags & PFSTATE_CREATEINPROG))
1262                         s = NULL;
1263         } else {
1264                 s = NULL;
1265         }
1266         if (globalstl)
1267                 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1268         return s;
1269 }
1270
1271 /* END state table stuff */
1272
1273 void
1274 pf_purge_thread(void *v)
1275 {
1276         globaldata_t save_gd = mycpu;
1277         int nloops = 0;
1278         int locked = 0;
1279         int nn;
1280         int endingit;
1281
1282         for (;;) {
1283                 tsleep(pf_purge_thread, PWAIT, "pftm", 1 * hz);
1284
1285                 endingit = pf_end_threads;
1286
1287                 for (nn = 0; nn < ncpus; ++nn) {
1288                         lwkt_setcpu_self(globaldata_find(nn));
1289
1290                         lwkt_gettoken(&pf_token);
1291                         lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1292                         crit_enter();
1293
1294                         /*
1295                          * process a fraction of the state table every second
1296                          */
1297                         if(!pf_purge_expired_states(
1298                                 1 + (pf_status.states /
1299                                      pf_default_rule.timeout[
1300                                         PFTM_INTERVAL]), 0)) {
1301                                 pf_purge_expired_states(
1302                                         1 + (pf_status.states /
1303                                              pf_default_rule.timeout[
1304                                                 PFTM_INTERVAL]), 1);
1305                         }
1306
1307                         /*
1308                          * purge other expired types every PFTM_INTERVAL
1309                          * seconds
1310                          */
1311                         if (++nloops >=
1312                             pf_default_rule.timeout[PFTM_INTERVAL]) {
1313                                 pf_purge_expired_fragments();
1314                                 if (!pf_purge_expired_src_nodes(locked)) {
1315                                         pf_purge_expired_src_nodes(1);
1316                                 }
1317                                 nloops = 0;
1318                         }
1319
1320                         /*
1321                          * If terminating the thread, clean everything out
1322                          * (on all cpus).
1323                          */
1324                         if (endingit) {
1325                                 pf_purge_expired_states(pf_status.states, 0);
1326                                 pf_purge_expired_fragments();
1327                                 pf_purge_expired_src_nodes(1);
1328                         }
1329
1330                         crit_exit();
1331                         lockmgr(&pf_consistency_lock, LK_RELEASE);
1332                         lwkt_reltoken(&pf_token);
1333                 }
1334                 lwkt_setcpu_self(save_gd);
1335                 if (endingit)
1336                         break;
1337         }
1338
1339         /*
1340          * Thread termination
1341          */
1342         pf_end_threads++;
1343         wakeup(pf_purge_thread);
1344         kthread_exit();
1345 }
1346
1347 u_int32_t
1348 pf_state_expires(const struct pf_state *state)
1349 {
1350         u_int32_t       timeout;
1351         u_int32_t       start;
1352         u_int32_t       end;
1353         u_int32_t       states;
1354
1355         /* handle all PFTM_* > PFTM_MAX here */
1356         if (state->timeout == PFTM_PURGE)
1357                 return (time_second);
1358         if (state->timeout == PFTM_UNTIL_PACKET)
1359                 return (0);
1360         KKASSERT(state->timeout != PFTM_UNLINKED);
1361         KKASSERT(state->timeout < PFTM_MAX);
1362         timeout = state->rule.ptr->timeout[state->timeout];
1363         if (!timeout)
1364                 timeout = pf_default_rule.timeout[state->timeout];
1365         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1366         if (start) {
1367                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1368                 states = state->rule.ptr->states_cur;
1369         } else {
1370                 start = pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1371                 end = pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1372                 states = pf_status.states;
1373         }
1374         if (end && states > start && start < end) {
1375                 if (states < end)
1376                         return (state->expire + timeout * (end - states) /
1377                             (end - start));
1378                 else
1379                         return (time_second);
1380         }
1381         return (state->expire + timeout);
1382 }
1383
1384 /*
1385  * (called with exclusive pf_token)
1386  */
1387 int
1388 pf_purge_expired_src_nodes(int waslocked)
1389 {
1390         struct pf_src_node *cur, *next;
1391         int locked = waslocked;
1392         int cpu = mycpu->gd_cpuid;
1393
1394         for (cur = RB_MIN(pf_src_tree, &tree_src_tracking[cpu]);
1395              cur;
1396              cur = next) {
1397                 next = RB_NEXT(pf_src_tree, &tree_src_tracking[cpu], cur);
1398
1399                 if (cur->states <= 0 && cur->expire <= time_second) {
1400                          if (!locked) {
1401                                  lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1402                                  next = RB_NEXT(pf_src_tree,
1403                                      &tree_src_tracking[cpu], cur);
1404                                  locked = 1;
1405                          }
1406                          if (cur->rule.ptr != NULL) {
1407                                 /*
1408                                  * decrements in rule should be ok, token is
1409                                  * held exclusively in this code path.
1410                                  */
1411                                  atomic_add_int(&cur->rule.ptr->src_nodes, -1);
1412                                  if (cur->rule.ptr->states_cur <= 0 &&
1413                                      cur->rule.ptr->max_src_nodes <= 0)
1414                                          pf_rm_rule(NULL, cur->rule.ptr);
1415                          }
1416                          RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], cur);
1417                          PF_INC_SCOUNTER(SCNT_SRC_NODE_REMOVALS);
1418                          atomic_add_int(&pf_status.src_nodes, -1);
1419                          kfree(cur, M_PFSRCTREEPL);
1420                 }
1421         }
1422         if (locked && !waslocked)
1423                 lockmgr(&pf_consistency_lock, LK_RELEASE);
1424         return(1);
1425 }
1426
1427 void
1428 pf_src_tree_remove_state(struct pf_state *s)
1429 {
1430         u_int32_t timeout;
1431
1432         if (s->src_node != NULL) {
1433                 if (s->src.tcp_est)
1434                         atomic_add_int(&s->src_node->conn, -1);
1435                 if (--s->src_node->states <= 0) {
1436                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1437                         if (!timeout) {
1438                                 timeout =
1439                                     pf_default_rule.timeout[PFTM_SRC_NODE];
1440                         }
1441                         s->src_node->expire = time_second + timeout;
1442                 }
1443         }
1444         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1445                 if (--s->nat_src_node->states <= 0) {
1446                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1447                         if (!timeout)
1448                                 timeout =
1449                                     pf_default_rule.timeout[PFTM_SRC_NODE];
1450                         s->nat_src_node->expire = time_second + timeout;
1451                 }
1452         }
1453         s->src_node = s->nat_src_node = NULL;
1454 }
1455
1456 /* callers should be at crit_enter() */
1457 void
1458 pf_unlink_state(struct pf_state *cur)
1459 {
1460         int cpu = mycpu->gd_cpuid;
1461
1462         if (cur->src.state == PF_TCPS_PROXY_DST) {
1463                 /* XXX wire key the right one? */
1464                 pf_send_tcp(cur->rule.ptr, cur->key[PF_SK_WIRE]->af,
1465                     &cur->key[PF_SK_WIRE]->addr[1],
1466                     &cur->key[PF_SK_WIRE]->addr[0],
1467                     cur->key[PF_SK_WIRE]->port[1],
1468                     cur->key[PF_SK_WIRE]->port[0],
1469                     cur->src.seqhi, cur->src.seqlo + 1,
1470                     TH_RST|TH_ACK, 0, 0, 0, 1, cur->tag, NULL, NULL);
1471         }
1472         RB_REMOVE(pf_state_tree_id, &tree_id[cpu], cur);
1473         if (cur->creatorid == pf_status.hostid)
1474                 pfsync_delete_state(cur);
1475         cur->timeout = PFTM_UNLINKED;
1476         pf_src_tree_remove_state(cur);
1477         pf_detach_state(cur);
1478 }
1479
1480 /*
1481  * callers should be at crit_enter() and hold pf_consistency_lock exclusively.
1482  * pf_token must also be held exclusively.
1483  */
1484 void
1485 pf_free_state(struct pf_state *cur)
1486 {
1487         int cpu = mycpu->gd_cpuid;
1488
1489         KKASSERT(cur->cpuid == cpu);
1490
1491         if (pfsyncif != NULL &&
1492             (pfsyncif->sc_bulk_send_next == cur ||
1493             pfsyncif->sc_bulk_terminator == cur))
1494                 return;
1495         KKASSERT(cur->timeout == PFTM_UNLINKED);
1496         /*
1497          * decrements in rule should be ok, token is
1498          * held exclusively in this code path.
1499          */
1500         if (--cur->rule.ptr->states_cur <= 0 &&
1501             cur->rule.ptr->src_nodes <= 0)
1502                 pf_rm_rule(NULL, cur->rule.ptr);
1503         if (cur->nat_rule.ptr != NULL) {
1504                 if (--cur->nat_rule.ptr->states_cur <= 0 &&
1505                         cur->nat_rule.ptr->src_nodes <= 0) {
1506                         pf_rm_rule(NULL, cur->nat_rule.ptr);
1507                 }
1508         }
1509         if (cur->anchor.ptr != NULL) {
1510                 if (--cur->anchor.ptr->states_cur <= 0)
1511                         pf_rm_rule(NULL, cur->anchor.ptr);
1512         }
1513         pf_normalize_tcp_cleanup(cur);
1514         pfi_kif_unref(cur->kif, PFI_KIF_REF_STATE);
1515
1516         /*
1517          * We may be freeing pf_purge_expired_states()'s saved scan entry,
1518          * adjust it if necessary.
1519          */
1520         if (purge_cur[cpu] == cur) {
1521                 kprintf("PURGE CONFLICT\n");
1522                 purge_cur[cpu] = TAILQ_NEXT(purge_cur[cpu], entry_list);
1523         }
1524         TAILQ_REMOVE(&state_list[cpu], cur, entry_list);
1525         if (cur->tag)
1526                 pf_tag_unref(cur->tag);
1527         kfree(cur, M_PFSTATEPL);
1528         PF_INC_FCOUNTER(FCNT_STATE_REMOVALS);
1529         atomic_add_int(&pf_status.states, -1);
1530 }
1531
1532 int
1533 pf_purge_expired_states(u_int32_t maxcheck, int waslocked)
1534 {
1535         struct pf_state         *cur;
1536         int locked = waslocked;
1537         int cpu = mycpu->gd_cpuid;
1538
1539         while (maxcheck--) {
1540                 /*
1541                  * Wrap to start of list when we hit the end
1542                  */
1543                 cur = purge_cur[cpu];
1544                 if (cur == NULL) {
1545                         cur = TAILQ_FIRST(&state_list[cpu]);
1546                         if (cur == NULL)
1547                                 break;  /* list empty */
1548                 }
1549
1550                 /*
1551                  * Setup next (purge_cur) while we process this one.  If
1552                  * we block and something else deletes purge_cur,
1553                  * pf_free_state() will adjust it further ahead.
1554                  */
1555                 purge_cur[cpu] = TAILQ_NEXT(cur, entry_list);
1556
1557                 if (cur->timeout == PFTM_UNLINKED) {
1558                         /* free unlinked state */
1559                         if (! locked) {
1560                                 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1561                                 locked = 1;
1562                         }
1563                         pf_free_state(cur);
1564                 } else if (pf_state_expires(cur) <= time_second) {
1565                         /* unlink and free expired state */
1566                         pf_unlink_state(cur);
1567                         if (! locked) {
1568                                 if (!lockmgr(&pf_consistency_lock, LK_EXCLUSIVE))
1569                                         return (0);
1570                                 locked = 1;
1571                         }
1572                         pf_free_state(cur);
1573                 }
1574         }
1575
1576         if (locked)
1577                 lockmgr(&pf_consistency_lock, LK_RELEASE);
1578         return (1);
1579 }
1580
1581 int
1582 pf_tbladdr_setup(struct pf_ruleset *rs, struct pf_addr_wrap *aw)
1583 {
1584         if (aw->type != PF_ADDR_TABLE)
1585                 return (0);
1586         if ((aw->p.tbl = pfr_attach_table(rs, aw->v.tblname)) == NULL)
1587                 return (1);
1588         return (0);
1589 }
1590
1591 void
1592 pf_tbladdr_remove(struct pf_addr_wrap *aw)
1593 {
1594         if (aw->type != PF_ADDR_TABLE || aw->p.tbl == NULL)
1595                 return;
1596         pfr_detach_table(aw->p.tbl);
1597         aw->p.tbl = NULL;
1598 }
1599
1600 void
1601 pf_tbladdr_copyout(struct pf_addr_wrap *aw)
1602 {
1603         struct pfr_ktable *kt = aw->p.tbl;
1604
1605         if (aw->type != PF_ADDR_TABLE || kt == NULL)
1606                 return;
1607         if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL)
1608                 kt = kt->pfrkt_root;
1609         aw->p.tbl = NULL;
1610         aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ?
1611                 kt->pfrkt_cnt : -1;
1612 }
1613
1614 void
1615 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1616 {
1617         switch (af) {
1618 #ifdef INET
1619         case AF_INET: {
1620                 u_int32_t a = ntohl(addr->addr32[0]);
1621                 kprintf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1622                     (a>>8)&255, a&255);
1623                 if (p) {
1624                         p = ntohs(p);
1625                         kprintf(":%u", p);
1626                 }
1627                 break;
1628         }
1629 #endif /* INET */
1630 #ifdef INET6
1631         case AF_INET6: {
1632                 u_int16_t b;
1633                 u_int8_t i, curstart, curend, maxstart, maxend;
1634                 curstart = curend = maxstart = maxend = 255;
1635                 for (i = 0; i < 8; i++) {
1636                         if (!addr->addr16[i]) {
1637                                 if (curstart == 255)
1638                                         curstart = i;
1639                                 curend = i;
1640                         } else {
1641                                 if ((curend - curstart) >
1642                                     (maxend - maxstart)) {
1643                                         maxstart = curstart;
1644                                         maxend = curend;
1645                                 }
1646                                 curstart = curend = 255;
1647                         }
1648                 }
1649                 if ((curend - curstart) >
1650                     (maxend - maxstart)) {
1651                         maxstart = curstart;
1652                         maxend = curend;
1653                 }
1654                 for (i = 0; i < 8; i++) {
1655                         if (i >= maxstart && i <= maxend) {
1656                                 if (i == 0)
1657                                         kprintf(":");
1658                                 if (i == maxend)
1659                                         kprintf(":");
1660                         } else {
1661                                 b = ntohs(addr->addr16[i]);
1662                                 kprintf("%x", b);
1663                                 if (i < 7)
1664                                         kprintf(":");
1665                         }
1666                 }
1667                 if (p) {
1668                         p = ntohs(p);
1669                         kprintf("[%u]", p);
1670                 }
1671                 break;
1672         }
1673 #endif /* INET6 */
1674         }
1675 }
1676
1677 void
1678 pf_print_state(struct pf_state *s)
1679 {
1680         pf_print_state_parts(s, NULL, NULL);
1681 }
1682
1683 void
1684 pf_print_state_parts(struct pf_state *s,
1685     struct pf_state_key *skwp, struct pf_state_key *sksp)
1686 {
1687         struct pf_state_key *skw, *sks;
1688         u_int8_t proto, dir;
1689
1690         /* Do our best to fill these, but they're skipped if NULL */
1691         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1692         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1693         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1694         dir = s ? s->direction : 0;
1695
1696         switch (proto) {
1697         case IPPROTO_TCP:
1698                 kprintf("TCP ");
1699                 break;
1700         case IPPROTO_UDP:
1701                 kprintf("UDP ");
1702                 break;
1703         case IPPROTO_ICMP:
1704                 kprintf("ICMP ");
1705                 break;
1706         case IPPROTO_ICMPV6:
1707                 kprintf("ICMPV6 ");
1708                 break;
1709         default:
1710                 kprintf("%u ", skw->proto);
1711                 break;
1712         }
1713         switch (dir) {
1714         case PF_IN:
1715                 kprintf(" in");
1716                 break;
1717         case PF_OUT:
1718                 kprintf(" out");
1719                 break;
1720         }
1721         if (skw) {
1722                 kprintf(" wire: ");
1723                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1724                 kprintf(" ");
1725                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1726         }
1727         if (sks) {
1728                 kprintf(" stack: ");
1729                 if (sks != skw) {
1730                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1731                         kprintf(" ");
1732                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1733                 } else
1734                         kprintf("-");
1735         }
1736         if (s) {
1737                 if (proto == IPPROTO_TCP) {
1738                         kprintf(" [lo=%u high=%u win=%u modulator=%u",
1739                             s->src.seqlo, s->src.seqhi,
1740                             s->src.max_win, s->src.seqdiff);
1741                         if (s->src.wscale && s->dst.wscale)
1742                                 kprintf(" wscale=%u",
1743                                     s->src.wscale & PF_WSCALE_MASK);
1744                         kprintf("]");
1745                         kprintf(" [lo=%u high=%u win=%u modulator=%u",
1746                             s->dst.seqlo, s->dst.seqhi,
1747                             s->dst.max_win, s->dst.seqdiff);
1748                         if (s->src.wscale && s->dst.wscale)
1749                                 kprintf(" wscale=%u",
1750                                 s->dst.wscale & PF_WSCALE_MASK);
1751                         kprintf("]");
1752                 }
1753                 kprintf(" %u:%u", s->src.state, s->dst.state);
1754         }
1755 }
1756
1757 void
1758 pf_print_flags(u_int8_t f)
1759 {
1760         if (f)
1761                 kprintf(" ");
1762         if (f & TH_FIN)
1763                 kprintf("F");
1764         if (f & TH_SYN)
1765                 kprintf("S");
1766         if (f & TH_RST)
1767                 kprintf("R");
1768         if (f & TH_PUSH)
1769                 kprintf("P");
1770         if (f & TH_ACK)
1771                 kprintf("A");
1772         if (f & TH_URG)
1773                 kprintf("U");
1774         if (f & TH_ECE)
1775                 kprintf("E");
1776         if (f & TH_CWR)
1777                 kprintf("W");
1778 }
1779
1780 #define PF_SET_SKIP_STEPS(i)                                    \
1781         do {                                                    \
1782                 while (head[i] != cur) {                        \
1783                         head[i]->skip[i].ptr = cur;             \
1784                         head[i] = TAILQ_NEXT(head[i], entries); \
1785                 }                                               \
1786         } while (0)
1787
1788 void
1789 pf_calc_skip_steps(struct pf_rulequeue *rules)
1790 {
1791         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1792         int i;
1793
1794         cur = TAILQ_FIRST(rules);
1795         prev = cur;
1796         for (i = 0; i < PF_SKIP_COUNT; ++i)
1797                 head[i] = cur;
1798         while (cur != NULL) {
1799
1800                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1801                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1802                 if (cur->direction != prev->direction)
1803                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1804                 if (cur->af != prev->af)
1805                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1806                 if (cur->proto != prev->proto)
1807                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1808                 if (cur->src.neg != prev->src.neg ||
1809                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1810                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1811                 if (cur->src.port[0] != prev->src.port[0] ||
1812                     cur->src.port[1] != prev->src.port[1] ||
1813                     cur->src.port_op != prev->src.port_op)
1814                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1815                 if (cur->dst.neg != prev->dst.neg ||
1816                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1817                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1818                 if (cur->dst.port[0] != prev->dst.port[0] ||
1819                     cur->dst.port[1] != prev->dst.port[1] ||
1820                     cur->dst.port_op != prev->dst.port_op)
1821                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1822
1823                 prev = cur;
1824                 cur = TAILQ_NEXT(cur, entries);
1825         }
1826         for (i = 0; i < PF_SKIP_COUNT; ++i)
1827                 PF_SET_SKIP_STEPS(i);
1828 }
1829
1830 int
1831 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1832 {
1833         if (aw1->type != aw2->type)
1834                 return (1);
1835         switch (aw1->type) {
1836         case PF_ADDR_ADDRMASK:
1837         case PF_ADDR_RANGE:
1838                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1839                         return (1);
1840                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1841                         return (1);
1842                 return (0);
1843         case PF_ADDR_DYNIFTL:
1844                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1845         case PF_ADDR_NOROUTE:
1846         case PF_ADDR_URPFFAILED:
1847                 return (0);
1848         case PF_ADDR_TABLE:
1849                 return (aw1->p.tbl != aw2->p.tbl);
1850         case PF_ADDR_RTLABEL:
1851                 return (aw1->v.rtlabel != aw2->v.rtlabel);
1852         default:
1853                 kprintf("invalid address type: %d\n", aw1->type);
1854                 return (1);
1855         }
1856 }
1857
1858 u_int16_t
1859 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1860 {
1861         u_int32_t       l;
1862
1863         if (udp && !cksum)
1864                 return (0x0000);
1865         l = cksum + old - new;
1866         l = (l >> 16) + (l & 65535);
1867         l = l & 65535;
1868         if (udp && !l)
1869                 return (0xFFFF);
1870         return (l);
1871 }
1872
1873 void
1874 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc,
1875     struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af)
1876 {
1877         struct pf_addr  ao;
1878         u_int16_t       po = *p;
1879
1880         PF_ACPY(&ao, a, af);
1881         PF_ACPY(a, an, af);
1882
1883         *p = pn;
1884
1885         switch (af) {
1886 #ifdef INET
1887         case AF_INET:
1888                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
1889                     ao.addr16[0], an->addr16[0], 0),
1890                     ao.addr16[1], an->addr16[1], 0);
1891                 *p = pn;
1892                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1893                     ao.addr16[0], an->addr16[0], u),
1894                     ao.addr16[1], an->addr16[1], u),
1895                     po, pn, u);
1896                 break;
1897 #endif /* INET */
1898 #ifdef INET6
1899         case AF_INET6:
1900                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1901                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1902                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1903                     ao.addr16[0], an->addr16[0], u),
1904                     ao.addr16[1], an->addr16[1], u),
1905                     ao.addr16[2], an->addr16[2], u),
1906                     ao.addr16[3], an->addr16[3], u),
1907                     ao.addr16[4], an->addr16[4], u),
1908                     ao.addr16[5], an->addr16[5], u),
1909                     ao.addr16[6], an->addr16[6], u),
1910                     ao.addr16[7], an->addr16[7], u),
1911                     po, pn, u);
1912                 break;
1913 #endif /* INET6 */
1914         }
1915 }
1916
1917
1918 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
1919 void
1920 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
1921 {
1922         u_int32_t       ao;
1923
1924         memcpy(&ao, a, sizeof(ao));
1925         memcpy(a, &an, sizeof(u_int32_t));
1926         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
1927             ao % 65536, an % 65536, u);
1928 }
1929
1930 #ifdef INET6
1931 void
1932 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
1933 {
1934         struct pf_addr  ao;
1935
1936         PF_ACPY(&ao, a, AF_INET6);
1937         PF_ACPY(a, an, AF_INET6);
1938
1939         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1940             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1941             pf_cksum_fixup(pf_cksum_fixup(*c,
1942             ao.addr16[0], an->addr16[0], u),
1943             ao.addr16[1], an->addr16[1], u),
1944             ao.addr16[2], an->addr16[2], u),
1945             ao.addr16[3], an->addr16[3], u),
1946             ao.addr16[4], an->addr16[4], u),
1947             ao.addr16[5], an->addr16[5], u),
1948             ao.addr16[6], an->addr16[6], u),
1949             ao.addr16[7], an->addr16[7], u);
1950 }
1951 #endif /* INET6 */
1952
1953 void
1954 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
1955     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
1956     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
1957 {
1958         struct pf_addr  oia, ooa;
1959
1960         PF_ACPY(&oia, ia, af);
1961         if (oa)
1962                 PF_ACPY(&ooa, oa, af);
1963
1964         /* Change inner protocol port, fix inner protocol checksum. */
1965         if (ip != NULL) {
1966                 u_int16_t       oip = *ip;
1967                 u_int32_t       opc = 0;
1968
1969                 if (pc != NULL)
1970                         opc = *pc;
1971                 *ip = np;
1972                 if (pc != NULL)
1973                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
1974                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
1975                 if (pc != NULL)
1976                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
1977         }
1978         /* Change inner ip address, fix inner ip and icmp checksums. */
1979         PF_ACPY(ia, na, af);
1980         switch (af) {
1981 #ifdef INET
1982         case AF_INET: {
1983                 u_int32_t        oh2c = *h2c;
1984
1985                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
1986                     oia.addr16[0], ia->addr16[0], 0),
1987                     oia.addr16[1], ia->addr16[1], 0);
1988                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
1989                     oia.addr16[0], ia->addr16[0], 0),
1990                     oia.addr16[1], ia->addr16[1], 0);
1991                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
1992                 break;
1993         }
1994 #endif /* INET */
1995 #ifdef INET6
1996         case AF_INET6:
1997                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1998                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1999                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2000                     oia.addr16[0], ia->addr16[0], u),
2001                     oia.addr16[1], ia->addr16[1], u),
2002                     oia.addr16[2], ia->addr16[2], u),
2003                     oia.addr16[3], ia->addr16[3], u),
2004                     oia.addr16[4], ia->addr16[4], u),
2005                     oia.addr16[5], ia->addr16[5], u),
2006                     oia.addr16[6], ia->addr16[6], u),
2007                     oia.addr16[7], ia->addr16[7], u);
2008                 break;
2009 #endif /* INET6 */
2010         }
2011         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2012         if (oa) {
2013                 PF_ACPY(oa, na, af);
2014                 switch (af) {
2015 #ifdef INET
2016                 case AF_INET:
2017                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2018                             ooa.addr16[0], oa->addr16[0], 0),
2019                             ooa.addr16[1], oa->addr16[1], 0);
2020                         break;
2021 #endif /* INET */
2022 #ifdef INET6
2023                 case AF_INET6:
2024                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2025                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2026                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2027                             ooa.addr16[0], oa->addr16[0], u),
2028                             ooa.addr16[1], oa->addr16[1], u),
2029                             ooa.addr16[2], oa->addr16[2], u),
2030                             ooa.addr16[3], oa->addr16[3], u),
2031                             ooa.addr16[4], oa->addr16[4], u),
2032                             ooa.addr16[5], oa->addr16[5], u),
2033                             ooa.addr16[6], oa->addr16[6], u),
2034                             ooa.addr16[7], oa->addr16[7], u);
2035                         break;
2036 #endif /* INET6 */
2037                 }
2038         }
2039 }
2040
2041
2042 /*
2043  * Need to modulate the sequence numbers in the TCP SACK option
2044  * (credits to Krzysztof Pfaff for report and patch)
2045  */
2046 int
2047 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2048     struct tcphdr *th, struct pf_state_peer *dst)
2049 {
2050         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2051         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2052         int copyback = 0, i, olen;
2053         struct raw_sackblock sack;
2054
2055 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2056         if (hlen < TCPOLEN_SACKLEN ||
2057             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2058                 return 0;
2059
2060         while (hlen >= TCPOLEN_SACKLEN) {
2061                 olen = opt[1];
2062                 switch (*opt) {
2063                 case TCPOPT_EOL:        /* FALLTHROUGH */
2064                 case TCPOPT_NOP:
2065                         opt++;
2066                         hlen--;
2067                         break;
2068                 case TCPOPT_SACK:
2069                         if (olen > hlen)
2070                                 olen = hlen;
2071                         if (olen >= TCPOLEN_SACKLEN) {
2072                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2073                                     i += TCPOLEN_SACK) {
2074                                         memcpy(&sack, &opt[i], sizeof(sack));
2075                                         pf_change_a(&sack.rblk_start, &th->th_sum,
2076                                             htonl(ntohl(sack.rblk_start) -
2077                                             dst->seqdiff), 0);
2078                                         pf_change_a(&sack.rblk_end, &th->th_sum,
2079                                             htonl(ntohl(sack.rblk_end) -
2080                                             dst->seqdiff), 0);
2081                                         memcpy(&opt[i], &sack, sizeof(sack));
2082                                 }
2083                                 copyback = 1;
2084                         }
2085                         /* FALLTHROUGH */
2086                 default:
2087                         if (olen < 2)
2088                                 olen = 2;
2089                         hlen -= olen;
2090                         opt += olen;
2091                 }
2092         }
2093
2094         if (copyback)
2095                 m_copyback(m, off + sizeof(*th), thoptlen, opts);
2096         return (copyback);
2097 }
2098
2099 void
2100 pf_send_tcp(const struct pf_rule *r, sa_family_t af,
2101     const struct pf_addr *saddr, const struct pf_addr *daddr,
2102     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2103     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2104     u_int16_t rtag, struct ether_header *eh, struct ifnet *ifp)
2105 {
2106         struct mbuf     *m;
2107         int              len = 0, tlen;
2108 #ifdef INET
2109         struct ip       *h = NULL;
2110 #endif /* INET */
2111 #ifdef INET6
2112         struct ip6_hdr  *h6 = NULL;
2113 #endif /* INET6 */
2114         struct tcphdr   *th = NULL;
2115         char            *opt;
2116
2117         ASSERT_LWKT_TOKEN_HELD(&pf_token);
2118
2119         /* maximum segment size tcp option */
2120         tlen = sizeof(struct tcphdr);
2121         if (mss)
2122                 tlen += 4;
2123
2124         switch (af) {
2125 #ifdef INET
2126         case AF_INET:
2127                 len = sizeof(struct ip) + tlen;
2128                 break;
2129 #endif /* INET */
2130 #ifdef INET6
2131         case AF_INET6:
2132                 len = sizeof(struct ip6_hdr) + tlen;
2133                 break;
2134 #endif /* INET6 */
2135         }
2136
2137         /*
2138          * Create outgoing mbuf.
2139          *
2140          * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
2141          * so make sure pf.flags is clear.
2142          */
2143         m = m_gethdr(M_NOWAIT, MT_HEADER);
2144         if (m == NULL) {
2145                 return;
2146         }
2147         if (tag)
2148                 m->m_pkthdr.fw_flags |= PF_MBUF_TAGGED;
2149         m->m_pkthdr.pf.flags = 0;
2150         m->m_pkthdr.pf.tag = rtag;
2151         /* XXX Recheck when upgrading to > 4.4 */
2152         m->m_pkthdr.pf.statekey = NULL;
2153         if (r != NULL && r->rtableid >= 0)
2154                 m->m_pkthdr.pf.rtableid = r->rtableid;
2155
2156 #ifdef ALTQ
2157         if (r != NULL && r->qid) {
2158                 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
2159                 m->m_pkthdr.pf.qid = r->qid;
2160                 m->m_pkthdr.pf.ecn_af = af;
2161                 m->m_pkthdr.pf.hdr = mtod(m, struct ip *);
2162         }
2163 #endif /* ALTQ */
2164         m->m_data += max_linkhdr;
2165         m->m_pkthdr.len = m->m_len = len;
2166         m->m_pkthdr.rcvif = NULL;
2167         bzero(m->m_data, len);
2168         switch (af) {
2169 #ifdef INET
2170         case AF_INET:
2171                 h = mtod(m, struct ip *);
2172
2173                 /* IP header fields included in the TCP checksum */
2174                 h->ip_p = IPPROTO_TCP;
2175                 h->ip_len = tlen;
2176                 h->ip_src.s_addr = saddr->v4.s_addr;
2177                 h->ip_dst.s_addr = daddr->v4.s_addr;
2178
2179                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2180                 break;
2181 #endif /* INET */
2182 #ifdef INET6
2183         case AF_INET6:
2184                 h6 = mtod(m, struct ip6_hdr *);
2185
2186                 /* IP header fields included in the TCP checksum */
2187                 h6->ip6_nxt = IPPROTO_TCP;
2188                 h6->ip6_plen = htons(tlen);
2189                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2190                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2191
2192                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2193                 break;
2194 #endif /* INET6 */
2195         }
2196
2197         /* TCP header */
2198         th->th_sport = sport;
2199         th->th_dport = dport;
2200         th->th_seq = htonl(seq);
2201         th->th_ack = htonl(ack);
2202         th->th_off = tlen >> 2;
2203         th->th_flags = flags;
2204         th->th_win = htons(win);
2205
2206         if (mss) {
2207                 opt = (char *)(th + 1);
2208                 opt[0] = TCPOPT_MAXSEG;
2209                 opt[1] = 4;
2210                 mss = htons(mss);
2211                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2212         }
2213
2214         switch (af) {
2215 #ifdef INET
2216         case AF_INET:
2217                 /* TCP checksum */
2218                 th->th_sum = in_cksum(m, len);
2219
2220                 /* Finish the IP header */
2221                 h->ip_v = 4;
2222                 h->ip_hl = sizeof(*h) >> 2;
2223                 h->ip_tos = IPTOS_LOWDELAY;
2224                 h->ip_len = len;
2225                 h->ip_off = path_mtu_discovery ? IP_DF : 0;
2226                 h->ip_ttl = ttl ? ttl : ip_defttl;
2227                 h->ip_sum = 0;
2228                 if (eh == NULL) {
2229                         lwkt_reltoken(&pf_token);
2230                         ip_output(m, NULL, NULL, 0, NULL, NULL);
2231                         lwkt_gettoken(&pf_token);
2232                 } else {
2233                         struct route             ro;
2234                         struct rtentry           rt;
2235                         struct ether_header     *e = (void *)ro.ro_dst.sa_data;
2236
2237                         if (ifp == NULL) {
2238                                 m_freem(m);
2239                                 return;
2240                         }
2241                         rt.rt_ifp = ifp;
2242                         ro.ro_rt = &rt;
2243                         ro.ro_dst.sa_len = sizeof(ro.ro_dst);
2244                         ro.ro_dst.sa_family = pseudo_AF_HDRCMPLT;
2245                         bcopy(eh->ether_dhost, e->ether_shost, ETHER_ADDR_LEN);
2246                         bcopy(eh->ether_shost, e->ether_dhost, ETHER_ADDR_LEN);
2247                         e->ether_type = eh->ether_type;
2248                         /* XXX_IMPORT: later */
2249                         lwkt_reltoken(&pf_token);
2250                         ip_output(m, NULL, &ro, 0, NULL, NULL);
2251                         lwkt_gettoken(&pf_token);
2252                 }
2253                 break;
2254 #endif /* INET */
2255 #ifdef INET6
2256         case AF_INET6:
2257                 /* TCP checksum */
2258                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2259                     sizeof(struct ip6_hdr), tlen);
2260
2261                 h6->ip6_vfc |= IPV6_VERSION;
2262                 h6->ip6_hlim = IPV6_DEFHLIM;
2263
2264                 lwkt_reltoken(&pf_token);
2265                 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
2266                 lwkt_gettoken(&pf_token);
2267                 break;
2268 #endif /* INET6 */
2269         }
2270 }
2271
2272 void
2273 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2274     struct pf_rule *r)
2275 {
2276         struct mbuf     *m0;
2277
2278         /*
2279          * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
2280          * so make sure pf.flags is clear.
2281          */
2282         if ((m0 = m_copy(m, 0, M_COPYALL)) == NULL)
2283                 return;
2284
2285         m0->m_pkthdr.fw_flags |= PF_MBUF_TAGGED;
2286         m0->m_pkthdr.pf.flags = 0;
2287         /* XXX Re-Check when Upgrading to > 4.4 */
2288         m0->m_pkthdr.pf.statekey = NULL;
2289
2290         if (r->rtableid >= 0)
2291                 m0->m_pkthdr.pf.rtableid = r->rtableid;
2292
2293 #ifdef ALTQ
2294         if (r->qid) {
2295                 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
2296                 m0->m_pkthdr.pf.qid = r->qid;
2297                 m0->m_pkthdr.pf.ecn_af = af;
2298                 m0->m_pkthdr.pf.hdr = mtod(m0, struct ip *);
2299         }
2300 #endif /* ALTQ */
2301
2302         switch (af) {
2303 #ifdef INET
2304         case AF_INET:
2305                 icmp_error(m0, type, code, 0, 0);
2306                 break;
2307 #endif /* INET */
2308 #ifdef INET6
2309         case AF_INET6:
2310                 icmp6_error(m0, type, code, 0);
2311                 break;
2312 #endif /* INET6 */
2313         }
2314 }
2315
2316 /*
2317  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2318  * If n is 0, they match if they are equal. If n is != 0, they match if they
2319  * are different.
2320  */
2321 int
2322 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2323     struct pf_addr *b, sa_family_t af)
2324 {
2325         int     match = 0;
2326
2327         switch (af) {
2328 #ifdef INET
2329         case AF_INET:
2330                 if ((a->addr32[0] & m->addr32[0]) ==
2331                     (b->addr32[0] & m->addr32[0]))
2332                         match++;
2333                 break;
2334 #endif /* INET */
2335 #ifdef INET6
2336         case AF_INET6:
2337                 if (((a->addr32[0] & m->addr32[0]) ==
2338                      (b->addr32[0] & m->addr32[0])) &&
2339                     ((a->addr32[1] & m->addr32[1]) ==
2340                      (b->addr32[1] & m->addr32[1])) &&
2341                     ((a->addr32[2] & m->addr32[2]) ==
2342                      (b->addr32[2] & m->addr32[2])) &&
2343                     ((a->addr32[3] & m->addr32[3]) ==
2344                      (b->addr32[3] & m->addr32[3])))
2345                         match++;
2346                 break;
2347 #endif /* INET6 */
2348         }
2349         if (match) {
2350                 if (n)
2351                         return (0);
2352                 else
2353                         return (1);
2354         } else {
2355                 if (n)
2356                         return (1);
2357                 else
2358                         return (0);
2359         }
2360 }
2361
2362 /*
2363  * Return 1 if b <= a <= e, otherwise return 0.
2364  */
2365 int
2366 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2367     struct pf_addr *a, sa_family_t af)
2368 {
2369         switch (af) {
2370 #ifdef INET
2371         case AF_INET:
2372                 if ((a->addr32[0] < b->addr32[0]) ||
2373                     (a->addr32[0] > e->addr32[0]))
2374                         return (0);
2375                 break;
2376 #endif /* INET */
2377 #ifdef INET6
2378         case AF_INET6: {
2379                 int     i;
2380
2381                 /* check a >= b */
2382                 for (i = 0; i < 4; ++i)
2383                         if (a->addr32[i] > b->addr32[i])
2384                                 break;
2385                         else if (a->addr32[i] < b->addr32[i])
2386                                 return (0);
2387                 /* check a <= e */
2388                 for (i = 0; i < 4; ++i)
2389                         if (a->addr32[i] < e->addr32[i])
2390                                 break;
2391                         else if (a->addr32[i] > e->addr32[i])
2392                                 return (0);
2393                 break;
2394         }
2395 #endif /* INET6 */
2396         }
2397         return (1);
2398 }
2399
2400 int
2401 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2402 {
2403         switch (op) {
2404         case PF_OP_IRG:
2405                 return ((p > a1) && (p < a2));
2406         case PF_OP_XRG:
2407                 return ((p < a1) || (p > a2));
2408         case PF_OP_RRG:
2409                 return ((p >= a1) && (p <= a2));
2410         case PF_OP_EQ:
2411                 return (p == a1);
2412         case PF_OP_NE:
2413                 return (p != a1);
2414         case PF_OP_LT:
2415                 return (p < a1);
2416         case PF_OP_LE:
2417                 return (p <= a1);
2418         case PF_OP_GT:
2419                 return (p > a1);
2420         case PF_OP_GE:
2421                 return (p >= a1);
2422         }
2423         return (0); /* never reached */
2424 }
2425
2426 int
2427 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2428 {
2429         a1 = ntohs(a1);
2430         a2 = ntohs(a2);
2431         p = ntohs(p);
2432         return (pf_match(op, a1, a2, p));
2433 }
2434
2435 int
2436 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2437 {
2438         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2439                 return (0);
2440         return (pf_match(op, a1, a2, u));
2441 }
2442
2443 int
2444 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2445 {
2446         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2447                 return (0);
2448         return (pf_match(op, a1, a2, g));
2449 }
2450
2451 int
2452 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag)
2453 {
2454         if (*tag == -1)
2455                 *tag = m->m_pkthdr.pf.tag;
2456
2457         return ((!r->match_tag_not && r->match_tag == *tag) ||
2458             (r->match_tag_not && r->match_tag != *tag));
2459 }
2460
2461 int
2462 pf_tag_packet(struct mbuf *m, int tag, int rtableid)
2463 {
2464         if (tag <= 0 && rtableid < 0)
2465                 return (0);
2466
2467         if (tag > 0)
2468                 m->m_pkthdr.pf.tag = tag;
2469         if (rtableid >= 0)
2470                 m->m_pkthdr.pf.rtableid = rtableid;
2471
2472         return (0);
2473 }
2474
2475 void
2476 pf_step_into_anchor(int *depth, struct pf_ruleset **rs, int n,
2477     struct pf_rule **r, struct pf_rule **a, int *match)
2478 {
2479         struct pf_anchor_stackframe     *f;
2480
2481         (*r)->anchor->match = 0;
2482         if (match)
2483                 *match = 0;
2484         if (*depth >= NELEM(pf_anchor_stack)) {
2485                 kprintf("pf_step_into_anchor: stack overflow\n");
2486                 *r = TAILQ_NEXT(*r, entries);
2487                 return;
2488         } else if (*depth == 0 && a != NULL)
2489                 *a = *r;
2490         f = pf_anchor_stack + (*depth)++;
2491         f->rs = *rs;
2492         f->r = *r;
2493         if ((*r)->anchor_wildcard) {
2494                 f->parent = &(*r)->anchor->children;
2495                 if ((f->child = RB_MIN(pf_anchor_node, f->parent)) ==
2496                     NULL) {
2497                         *r = NULL;
2498                         return;
2499                 }
2500                 *rs = &f->child->ruleset;
2501         } else {
2502                 f->parent = NULL;
2503                 f->child = NULL;
2504                 *rs = &(*r)->anchor->ruleset;
2505         }
2506         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2507 }
2508
2509 int
2510 pf_step_out_of_anchor(int *depth, struct pf_ruleset **rs, int n,
2511     struct pf_rule **r, struct pf_rule **a, int *match)
2512 {
2513         struct pf_anchor_stackframe     *f;
2514         int quick = 0;
2515
2516         do {
2517                 if (*depth <= 0)
2518                         break;
2519                 f = pf_anchor_stack + *depth - 1;
2520                 if (f->parent != NULL && f->child != NULL) {
2521                         if (f->child->match ||
2522                             (match != NULL && *match)) {
2523                                 f->r->anchor->match = 1;
2524                                 *match = 0;
2525                         }
2526                         f->child = RB_NEXT(pf_anchor_node, f->parent, f->child);
2527                         if (f->child != NULL) {
2528                                 *rs = &f->child->ruleset;
2529                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2530                                 if (*r == NULL)
2531                                         continue;
2532                                 else
2533                                         break;
2534                         }
2535                 }
2536                 (*depth)--;
2537                 if (*depth == 0 && a != NULL)
2538                         *a = NULL;
2539                 *rs = f->rs;
2540                 if (f->r->anchor->match || (match != NULL && *match))
2541                         quick = f->r->quick;
2542                 *r = TAILQ_NEXT(f->r, entries);
2543         } while (*r == NULL);
2544
2545         return (quick);
2546 }
2547
2548 #ifdef INET6
2549 void
2550 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2551     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2552 {
2553         switch (af) {
2554 #ifdef INET
2555         case AF_INET:
2556                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2557                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2558                 break;
2559 #endif /* INET */
2560         case AF_INET6:
2561                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2562                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2563                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2564                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2565                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2566                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2567                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2568                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2569                 break;
2570         }
2571 }
2572
2573 void
2574 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2575 {
2576         switch (af) {
2577 #ifdef INET
2578         case AF_INET:
2579                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2580                 break;
2581 #endif /* INET */
2582         case AF_INET6:
2583                 if (addr->addr32[3] == 0xffffffff) {
2584                         addr->addr32[3] = 0;
2585                         if (addr->addr32[2] == 0xffffffff) {
2586                                 addr->addr32[2] = 0;
2587                                 if (addr->addr32[1] == 0xffffffff) {
2588                                         addr->addr32[1] = 0;
2589                                         addr->addr32[0] =
2590                                             htonl(ntohl(addr->addr32[0]) + 1);
2591                                 } else
2592                                         addr->addr32[1] =
2593                                             htonl(ntohl(addr->addr32[1]) + 1);
2594                         } else
2595                                 addr->addr32[2] =
2596                                     htonl(ntohl(addr->addr32[2]) + 1);
2597                 } else
2598                         addr->addr32[3] =
2599                             htonl(ntohl(addr->addr32[3]) + 1);
2600                 break;
2601         }
2602 }
2603 #endif /* INET6 */
2604
2605 #define mix(a,b,c) \
2606         do {                                    \
2607                 a -= b; a -= c; a ^= (c >> 13); \
2608                 b -= c; b -= a; b ^= (a << 8);  \
2609                 c -= a; c -= b; c ^= (b >> 13); \
2610                 a -= b; a -= c; a ^= (c >> 12); \
2611                 b -= c; b -= a; b ^= (a << 16); \
2612                 c -= a; c -= b; c ^= (b >> 5);  \
2613                 a -= b; a -= c; a ^= (c >> 3);  \
2614                 b -= c; b -= a; b ^= (a << 10); \
2615                 c -= a; c -= b; c ^= (b >> 15); \
2616         } while (0)
2617
2618 /*
2619  * hash function based on bridge_hash in if_bridge.c
2620  */
2621 void
2622 pf_hash(struct pf_addr *inaddr, struct pf_addr *hash,
2623     struct pf_poolhashkey *key, sa_family_t af)
2624 {
2625         u_int32_t       a = 0x9e3779b9, b = 0x9e3779b9, c = key->key32[0];
2626
2627         switch (af) {
2628 #ifdef INET
2629         case AF_INET:
2630                 a += inaddr->addr32[0];
2631                 b += key->key32[1];
2632                 mix(a, b, c);
2633                 hash->addr32[0] = c + key->key32[2];
2634                 break;
2635 #endif /* INET */
2636 #ifdef INET6
2637         case AF_INET6:
2638                 a += inaddr->addr32[0];
2639                 b += inaddr->addr32[2];
2640                 mix(a, b, c);
2641                 hash->addr32[0] = c;
2642                 a += inaddr->addr32[1];
2643                 b += inaddr->addr32[3];
2644                 c += key->key32[1];
2645                 mix(a, b, c);
2646                 hash->addr32[1] = c;
2647                 a += inaddr->addr32[2];
2648                 b += inaddr->addr32[1];
2649                 c += key->key32[2];
2650                 mix(a, b, c);
2651                 hash->addr32[2] = c;
2652                 a += inaddr->addr32[3];
2653                 b += inaddr->addr32[0];
2654                 c += key->key32[3];
2655                 mix(a, b, c);
2656                 hash->addr32[3] = c;
2657                 break;
2658 #endif /* INET6 */
2659         }
2660 }
2661
2662 int
2663 pf_map_addr(sa_family_t af, struct pf_rule *r, struct pf_addr *saddr,
2664     struct pf_addr *naddr, struct pf_addr *init_addr, struct pf_src_node **sn)
2665 {
2666         unsigned char            hash[16];
2667         struct pf_pool          *rpool = &r->rpool;
2668         struct pf_pooladdr      *acur = rpool->cur;
2669         struct pf_pooladdr      *cur;
2670         struct pf_addr          *raddr;
2671         struct pf_addr          *rmask;
2672         struct pf_addr          counter;
2673         struct pf_src_node       k;
2674         int cpu = mycpu->gd_cpuid;
2675         int tblidx;
2676
2677         bzero(hash, sizeof(hash));      /* avoid gcc warnings */
2678
2679         /*
2680          * NOTE! rpool->cur and rpool->tblidx can be iterators and thus
2681          *       may represent a SMP race due to the shared nature of the
2682          *       rpool structure.  We allow the race and ensure that updates
2683          *       do not create a fatal condition.
2684          */
2685         cpu_ccfence();
2686         cur = acur;
2687         raddr = &cur->addr.v.a.addr;
2688         rmask = &cur->addr.v.a.mask;
2689
2690         if (*sn == NULL && r->rpool.opts & PF_POOL_STICKYADDR &&
2691             (r->rpool.opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
2692                 k.af = af;
2693                 PF_ACPY(&k.addr, saddr, af);
2694                 if (r->rule_flag & PFRULE_RULESRCTRACK ||
2695                     r->rpool.opts & PF_POOL_STICKYADDR)
2696                         k.rule.ptr = r;
2697                 else
2698                         k.rule.ptr = NULL;
2699                 PF_INC_SCOUNTER(SCNT_SRC_NODE_SEARCH);
2700                 *sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k);
2701                 if (*sn != NULL && !PF_AZERO(&(*sn)->raddr, af)) {
2702                         PF_ACPY(naddr, &(*sn)->raddr, af);
2703                         if (pf_status.debug >= PF_DEBUG_MISC) {
2704                                 kprintf("pf_map_addr: src tracking maps ");
2705                                 pf_print_host(&k.addr, 0, af);
2706                                 kprintf(" to ");
2707                                 pf_print_host(naddr, 0, af);
2708                                 kprintf("\n");
2709                         }
2710                         return (0);
2711                 }
2712         }
2713
2714         if (cur->addr.type == PF_ADDR_NOROUTE)
2715                 return (1);
2716         if (cur->addr.type == PF_ADDR_DYNIFTL) {
2717                 switch (af) {
2718 #ifdef INET
2719                 case AF_INET:
2720                         if (cur->addr.p.dyn->pfid_acnt4 < 1 &&
2721                             (rpool->opts & PF_POOL_TYPEMASK) !=
2722                             PF_POOL_ROUNDROBIN)
2723                                 return (1);
2724                         raddr = &cur->addr.p.dyn->pfid_addr4;
2725                         rmask = &cur->addr.p.dyn->pfid_mask4;
2726                         break;
2727 #endif /* INET */
2728 #ifdef INET6
2729                 case AF_INET6:
2730                         if (cur->addr.p.dyn->pfid_acnt6 < 1 &&
2731                             (rpool->opts & PF_POOL_TYPEMASK) !=
2732                             PF_POOL_ROUNDROBIN)
2733                                 return (1);
2734                         raddr = &cur->addr.p.dyn->pfid_addr6;
2735                         rmask = &cur->addr.p.dyn->pfid_mask6;
2736                         break;
2737 #endif /* INET6 */
2738                 }
2739         } else if (cur->addr.type == PF_ADDR_TABLE) {
2740                 if ((rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_ROUNDROBIN)
2741                         return (1); /* unsupported */
2742         } else {
2743                 raddr = &cur->addr.v.a.addr;
2744                 rmask = &cur->addr.v.a.mask;
2745         }
2746
2747         switch (rpool->opts & PF_POOL_TYPEMASK) {
2748         case PF_POOL_NONE:
2749                 PF_ACPY(naddr, raddr, af);
2750                 break;
2751         case PF_POOL_BITMASK:
2752                 PF_POOLMASK(naddr, raddr, rmask, saddr, af);
2753                 break;
2754         case PF_POOL_RANDOM:
2755                 if (init_addr != NULL && PF_AZERO(init_addr, af)) {
2756                         switch (af) {
2757 #ifdef INET
2758                         case AF_INET:
2759                                 counter.addr32[0] = htonl(karc4random());
2760                                 break;
2761 #endif /* INET */
2762 #ifdef INET6
2763                         case AF_INET6:
2764                                 if (rmask->addr32[3] != 0xffffffff)
2765                                         counter.addr32[3] =
2766                                                 htonl(karc4random());
2767                                 else
2768                                         break;
2769                                 if (rmask->addr32[2] != 0xffffffff)
2770                                         counter.addr32[2] =
2771                                                 htonl(karc4random());
2772                                 else
2773                                         break;
2774                                 if (rmask->addr32[1] != 0xffffffff)
2775                                         counter.addr32[1] =
2776                                                 htonl(karc4random());
2777                                 else
2778                                         break;
2779                                 if (rmask->addr32[0] != 0xffffffff)
2780                                         counter.addr32[0] =
2781                                                 htonl(karc4random());
2782                                 break;
2783 #endif /* INET6 */
2784                         }
2785                         PF_POOLMASK(naddr, raddr, rmask, &counter, af);
2786                         PF_ACPY(init_addr, naddr, af);
2787
2788                 } else {
2789                         counter = rpool->counter;
2790                         cpu_ccfence();
2791                         PF_AINC(&counter, af);
2792                         PF_POOLMASK(naddr, raddr, rmask, &counter, af);
2793                         rpool->counter = counter;
2794                 }
2795                 break;
2796         case PF_POOL_SRCHASH:
2797                 pf_hash(saddr, (struct pf_addr *)&hash, &rpool->key, af);
2798                 PF_POOLMASK(naddr, raddr, rmask, (struct pf_addr *)&hash, af);
2799                 break;
2800         case PF_POOL_ROUNDROBIN:
2801                 tblidx = rpool->tblidx;
2802                 counter = rpool->counter;
2803                 if (cur->addr.type == PF_ADDR_TABLE) {
2804                         if (!pfr_pool_get(cur->addr.p.tbl,
2805                             &tblidx, &counter,
2806                             &raddr, &rmask, af)) {
2807                                 goto get_addr;
2808                         }
2809                 } else if (cur->addr.type == PF_ADDR_DYNIFTL) {
2810                         if (!pfr_pool_get(cur->addr.p.dyn->pfid_kt,
2811                             &tblidx, &counter,
2812                             &raddr, &rmask, af)) {
2813                                 goto get_addr;
2814                         }
2815                 } else if (pf_match_addr(0, raddr, rmask,
2816                                          &counter, af)) {
2817                         goto get_addr;
2818                 }
2819
2820         try_next:
2821                 if ((cur = TAILQ_NEXT(cur, entries)) == NULL)
2822                         cur = TAILQ_FIRST(&rpool->list);
2823                 if (cur->addr.type == PF_ADDR_TABLE) {
2824                         tblidx = -1;
2825                         if (pfr_pool_get(cur->addr.p.tbl,
2826                             &tblidx, &counter,
2827                             &raddr, &rmask, af)) {
2828                                 /* table contains no address of type 'af' */
2829                                 if (cur != acur)
2830                                         goto try_next;
2831                                 return (1);
2832                         }
2833                 } else if (cur->addr.type == PF_ADDR_DYNIFTL) {
2834                         tblidx = -1;
2835                         if (pfr_pool_get(cur->addr.p.dyn->pfid_kt,
2836                             &tblidx, &counter,
2837                             &raddr, &rmask, af)) {
2838                                 /* table contains no address of type 'af' */
2839                                 if (cur != acur)
2840                                         goto try_next;
2841                                 return (1);
2842                         }
2843                 } else {
2844                         raddr = &cur->addr.v.a.addr;
2845                         rmask = &cur->addr.v.a.mask;
2846                         PF_ACPY(&counter, raddr, af);
2847                 }
2848
2849         get_addr:
2850                 rpool->cur = cur;
2851                 rpool->tblidx = tblidx;
2852                 PF_ACPY(naddr, &counter, af);
2853                 if (init_addr != NULL && PF_AZERO(init_addr, af))
2854                         PF_ACPY(init_addr, naddr, af);
2855                 PF_AINC(&counter, af);
2856                 rpool->counter = counter;
2857                 break;
2858         }
2859         if (*sn != NULL)
2860                 PF_ACPY(&(*sn)->raddr, naddr, af);
2861
2862         if (pf_status.debug >= PF_DEBUG_MISC &&
2863             (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
2864                 kprintf("pf_map_addr: selected address ");
2865                 pf_print_host(naddr, 0, af);
2866                 kprintf("\n");
2867         }
2868
2869         return (0);
2870 }
2871
2872 int
2873 pf_get_sport(struct pf_pdesc *pd, sa_family_t af,
2874              u_int8_t proto, struct pf_rule *r,
2875              struct pf_addr *saddr, struct pf_addr *daddr,
2876              u_int16_t sport, u_int16_t dport,
2877              struct pf_addr *naddr, u_int16_t *nport,
2878              u_int16_t low, u_int16_t high, struct pf_src_node **sn)
2879 {
2880         struct pf_state_key_cmp key;
2881         struct pf_addr          init_addr;
2882         u_int16_t               cut;
2883         u_int32_t               hash_base = 0;
2884         int                     do_hash = 0;
2885
2886         bzero(&init_addr, sizeof(init_addr));
2887         if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
2888                 return (1);
2889
2890         if (proto == IPPROTO_ICMP) {
2891                 low = 1;
2892                 high = 65535;
2893         }
2894
2895         bzero(&key, sizeof(key));
2896         key.af = af;
2897         key.proto = proto;
2898         key.port[0] = dport;
2899         PF_ACPY(&key.addr[0], daddr, key.af);
2900
2901         do {
2902                 PF_ACPY(&key.addr[1], naddr, key.af);
2903
2904                 /*
2905                  * We want to select a port that calculates to a toeplitz hash
2906                  * that masks to the same cpu, otherwise the response may
2907                  * not see the new state.
2908                  *
2909                  * We can still do this even if the kernel is disregarding
2910                  * the hash and vectoring the packets to a specific cpu,
2911                  * but it will reduce the number of ports we can use.
2912                  */
2913                 switch(af) {
2914                 case AF_INET:
2915                         if (proto == IPPROTO_TCP) {
2916                                 do_hash = 1;
2917                                 hash_base = toeplitz_piecemeal_port(dport) ^
2918                                     toeplitz_piecemeal_addr(daddr->v4.s_addr) ^
2919                                     toeplitz_piecemeal_addr(naddr->v4.s_addr);
2920                         }
2921                         break;
2922                 case AF_INET6:
2923                         /* XXX TODO XXX */
2924                 default:
2925                         /* XXX TODO XXX */
2926                         break;
2927                 }
2928
2929                 /*
2930                  * port search; start random, step;
2931                  * similar 2 portloop in in_pcbbind
2932                  *
2933                  * WARNING! We try to match such that the kernel will
2934                  *          dispatch the translated host/port to the same
2935                  *          cpu, but this might not be possible.
2936                  *
2937                  *          In the case where the port is fixed, or for the
2938                  *          UDP case (whos toeplitz does not incorporate the
2939                  *          port), we set not_cpu_localized which ultimately
2940                  *          causes the pf_state_tree element
2941                  *
2942                  * XXX fixed ports present a problem for cpu localization.
2943                  */
2944                 if (!(proto == IPPROTO_TCP ||
2945                       proto == IPPROTO_UDP ||
2946                       proto == IPPROTO_ICMP)) {
2947                         /*
2948                          * non-specific protocol, leave port intact.
2949                          */
2950                         key.port[1] = sport;
2951                         if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2952                                 *nport = sport;
2953                                 pd->not_cpu_localized = 1;
2954                                 return (0);
2955                         }
2956                 } else if (low == 0 && high == 0) {
2957                         /*
2958                          * static-port same as originator.
2959                          */
2960                         key.port[1] = sport;
2961                         if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2962                                 *nport = sport;
2963                                 pd->not_cpu_localized = 1;
2964                                 return (0);
2965                         }
2966                 } else if (low == high) {
2967                         /*
2968                          * specific port as specified.
2969                          */
2970                         key.port[1] = htons(low);
2971                         if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2972                                 *nport = htons(low);
2973                                 pd->not_cpu_localized = 1;
2974                                 return (0);
2975                         }
2976                 } else {
2977                         /*
2978                          * normal dynamic port
2979                          */
2980                         u_int16_t tmp;
2981
2982                         if (low > high) {
2983                                 tmp = low;
2984                                 low = high;
2985                                 high = tmp;
2986                         }
2987                         /* low < high */
2988                         cut = htonl(karc4random()) % (1 + high - low) + low;
2989                         /* low <= cut <= high */
2990                         for (tmp = cut; tmp <= high; ++(tmp)) {
2991                                 key.port[1] = htons(tmp);
2992                                 if (do_hash) {
2993                                         uint32_t hash;
2994
2995                                         hash = hash_base ^
2996                                         toeplitz_piecemeal_port(key.port[1]);
2997                                         if (netisr_hashcpu(hash) != mycpuid)
2998                                                 continue;
2999                                 }
3000                                 if (pf_find_state_all(&key, PF_IN, NULL) ==
3001                                     NULL && !in_baddynamic(tmp, proto)) {
3002                                         if (proto == IPPROTO_UDP)
3003                                                 pd->not_cpu_localized = 1;
3004                                         *nport = htons(tmp);
3005                                         return (0);
3006                                 }
3007                         }
3008                         for (tmp = cut - 1; tmp >= low; --(tmp)) {
3009                                 key.port[1] = htons(tmp);
3010                                 if (do_hash) {
3011                                         uint32_t hash;
3012
3013                                         hash = hash_base ^
3014                                         toeplitz_piecemeal_port(key.port[1]);
3015                                         if (netisr_hashcpu(hash) != mycpuid)
3016                                                 continue;
3017                                 }
3018                                 if (pf_find_state_all(&key, PF_IN, NULL) ==
3019                                     NULL && !in_baddynamic(tmp, proto)) {
3020                                         if (proto == IPPROTO_UDP)
3021                                                 pd->not_cpu_localized = 1;
3022                                         *nport = htons(tmp);
3023                                         return (0);
3024                                 }
3025                         }
3026                 }
3027
3028                 /*
3029                  * Next address
3030                  */
3031                 switch (r->rpool.opts & PF_POOL_TYPEMASK) {
3032                 case PF_POOL_RANDOM:
3033                 case PF_POOL_ROUNDROBIN:
3034                         if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
3035                                 return (1);
3036                         break;
3037                 case PF_POOL_NONE:
3038                 case PF_POOL_SRCHASH:
3039                 case PF_POOL_BITMASK:
3040                 default:
3041                         return (1);
3042                 }
3043         } while (! PF_AEQ(&init_addr, naddr, af) );
3044         return (1);                                     /* none available */
3045 }
3046
3047 struct pf_rule *
3048 pf_match_translation(struct pf_pdesc *pd, struct mbuf *m, int off,
3049     int direction, struct pfi_kif *kif, struct pf_addr *saddr, u_int16_t sport,
3050     struct pf_addr *daddr, u_int16_t dport, int rs_num)
3051 {
3052         struct pf_rule          *r, *rm = NULL;
3053         struct pf_ruleset       *ruleset = NULL;
3054         int                      tag = -1;
3055         int                      rtableid = -1;
3056         int                      asd = 0;
3057
3058         r = TAILQ_FIRST(pf_main_ruleset.rules[rs_num].active.ptr);
3059         while (r && rm == NULL) {
3060                 struct pf_rule_addr     *src = NULL, *dst = NULL;
3061                 struct pf_addr_wrap     *xdst = NULL;
3062                 struct pf_pooladdr      *cur;
3063
3064                 if (r->action == PF_BINAT && direction == PF_IN) {
3065                         src = &r->dst;
3066                         cur = r->rpool.cur;     /* SMP race possible */
3067                         cpu_ccfence();
3068                         if (cur)
3069                                 xdst = &cur->addr;
3070                 } else {
3071                         src = &r->src;
3072                         dst = &r->dst;
3073                 }
3074
3075                 r->evaluations++;
3076                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3077                         r = r->skip[PF_SKIP_IFP].ptr;
3078                 else if (r->direction && r->direction != direction)
3079                         r = r->skip[PF_SKIP_DIR].ptr;
3080                 else if (r->af && r->af != pd->af)
3081                         r = r->skip[PF_SKIP_AF].ptr;
3082                 else if (r->proto && r->proto != pd->proto)
3083                         r = r->skip[PF_SKIP_PROTO].ptr;
3084                 else if (PF_MISMATCHAW(&src->addr, saddr, pd->af,
3085                     src->neg, kif))
3086                         r = r->skip[src == &r->src ? PF_SKIP_SRC_ADDR :
3087                             PF_SKIP_DST_ADDR].ptr;
3088                 else if (src->port_op && !pf_match_port(src->port_op,
3089                     src->port[0], src->port[1], sport))
3090                         r = r->skip[src == &r->src ? PF_SKIP_SRC_PORT :
3091                             PF_SKIP_DST_PORT].ptr;
3092                 else if (dst != NULL &&
3093                     PF_MISMATCHAW(&dst->addr, daddr, pd->af, dst->neg, NULL))
3094                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3095                 else if (xdst != NULL && PF_MISMATCHAW(xdst, daddr, pd->af,
3096                     0, NULL))
3097                         r = TAILQ_NEXT(r, entries);
3098                 else if (dst != NULL && dst->port_op &&
3099                     !pf_match_port(dst->port_op, dst->port[0],
3100                     dst->port[1], dport))
3101                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3102                 else if (r->match_tag && !pf_match_tag(m, r, &tag))
3103                         r = TAILQ_NEXT(r, entries);
3104                 else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto !=
3105                     IPPROTO_TCP || !pf_osfp_match(pf_osfp_fingerprint(pd, m,
3106                     off, pd->hdr.tcp), r->os_fingerprint)))
3107                         r = TAILQ_NEXT(r, entries);
3108                 else {
3109                         if (r->tag)
3110                                 tag = r->tag;
3111                         if (r->rtableid >= 0)
3112                                 rtableid = r->rtableid;
3113                         if (r->anchor == NULL) {
3114                                 rm = r;
3115                         } else
3116                                 pf_step_into_anchor(&asd, &ruleset, rs_num,
3117                                     &r, NULL, NULL);
3118                 }
3119                 if (r == NULL)
3120                         pf_step_out_of_anchor(&asd, &ruleset, rs_num, &r,
3121                             NULL, NULL);
3122         }
3123         if (pf_tag_packet(m, tag, rtableid))
3124                 return (NULL);
3125         if (rm != NULL && (rm->action == PF_NONAT ||
3126             rm->action == PF_NORDR || rm->action == PF_NOBINAT))
3127                 return (NULL);
3128         return (rm);
3129 }
3130
3131 struct pf_rule *
3132 pf_get_translation(struct pf_pdesc *pd, struct mbuf *m, int off, int direction,
3133     struct pfi_kif *kif, struct pf_src_node **sn,
3134     struct pf_state_key **skw, struct pf_state_key **sks,
3135     struct pf_state_key **skp, struct pf_state_key **nkp,
3136     struct pf_addr *saddr, struct pf_addr *daddr,
3137     u_int16_t sport, u_int16_t dport)
3138 {
3139         struct pf_rule  *r = NULL;
3140
3141         if (direction == PF_OUT) {
3142                 r = pf_match_translation(pd, m, off, direction, kif, saddr,
3143                     sport, daddr, dport, PF_RULESET_BINAT);
3144                 if (r == NULL)
3145                         r = pf_match_translation(pd, m, off, direction, kif,
3146                             saddr, sport, daddr, dport, PF_RULESET_NAT);
3147         } else {
3148                 r = pf_match_translation(pd, m, off, direction, kif, saddr,
3149                     sport, daddr, dport, PF_RULESET_RDR);
3150                 if (r == NULL)
3151                         r = pf_match_translation(pd, m, off, direction, kif,
3152                             saddr, sport, daddr, dport, PF_RULESET_BINAT);
3153         }
3154
3155         if (r != NULL) {
3156                 struct pf_addr  *naddr;
3157                 u_int16_t       *nport;
3158
3159                 if (pf_state_key_setup(pd, r, skw, sks, skp, nkp,
3160                     saddr, daddr, sport, dport))
3161                         return r;
3162
3163                 /* XXX We only modify one side for now. */
3164                 naddr = &(*nkp)->addr[1];
3165                 nport = &(*nkp)->port[1];
3166
3167                 /*
3168                  * NOTE: Currently all translations will clear
3169                  *       BRIDGE_MBUF_TAGGED, telling the bridge to
3170                  *       ignore the original input encapsulation.
3171                  */
3172                 switch (r->action) {
3173                 case PF_NONAT:
3174                 case PF_NOBINAT:
3175                 case PF_NORDR:
3176                         return (NULL);
3177                 case PF_NAT:
3178                         m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3179                         if (pf_get_sport(pd, pd->af, pd->proto, r,
3180                             saddr, daddr, sport, dport,
3181                             naddr, nport, r->rpool.proxy_port[0],
3182                             r->rpool.proxy_port[1], sn)) {
3183                                 DPFPRINTF(PF_DEBUG_MISC,
3184                                     ("pf: NAT proxy port allocation "
3185                                     "(%u-%u) failed\n",
3186                                     r->rpool.proxy_port[0],
3187                                     r->rpool.proxy_port[1]));
3188                                 return (NULL);
3189                         }
3190                         break;
3191                 case PF_BINAT:
3192                         m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3193                         switch (direction) {
3194                         case PF_OUT:
3195                                 if (r->rpool.cur->addr.type == PF_ADDR_DYNIFTL){
3196                                         switch (pd->af) {
3197 #ifdef INET
3198                                         case AF_INET:
3199                                                 if (r->rpool.cur->addr.p.dyn->
3200                                                     pfid_acnt4 < 1)
3201                                                         return (NULL);
3202                                                 PF_POOLMASK(naddr,
3203                                                     &r->rpool.cur->addr.p.dyn->
3204                                                     pfid_addr4,
3205                                                     &r->rpool.cur->addr.p.dyn->
3206                                                     pfid_mask4,
3207                                                     saddr, AF_INET);
3208                                                 break;
3209 #endif /* INET */
3210 #ifdef INET6
3211                                         case AF_INET6:
3212                                                 if (r->rpool.cur->addr.p.dyn->
3213                                                     pfid_acnt6 < 1)
3214                                                         return (NULL);
3215                                                 PF_POOLMASK(naddr,
3216                                                     &r->rpool.cur->addr.p.dyn->
3217                                                     pfid_addr6,
3218                                                     &r->rpool.cur->addr.p.dyn->
3219                                                     pfid_mask6,
3220                                                     saddr, AF_INET6);
3221                                                 break;
3222 #endif /* INET6 */
3223                                         }
3224                                 } else
3225                                         PF_POOLMASK(naddr,
3226                                             &r->rpool.cur->addr.v.a.addr,
3227                                             &r->rpool.cur->addr.v.a.mask,
3228                                             saddr, pd->af);
3229                                 break;
3230                         case PF_IN:
3231                                 if (r->src.addr.type == PF_ADDR_DYNIFTL) {
3232                                         switch (pd->af) {
3233 #ifdef INET
3234                                         case AF_INET:
3235                                                 if (r->src.addr.p.dyn->
3236                                                     pfid_acnt4 < 1)
3237                                                         return (NULL);
3238                                                 PF_POOLMASK(naddr,
3239                                                     &r->src.addr.p.dyn->
3240                                                     pfid_addr4,
3241                                                     &r->src.addr.p.dyn->
3242                                                     pfid_mask4,
3243                                                     daddr, AF_INET);
3244                                                 break;
3245 #endif /* INET */
3246 #ifdef INET6
3247                                         case AF_INET6:
3248                                                 if (r->src.addr.p.dyn->
3249                                                     pfid_acnt6 < 1)
3250                                                         return (NULL);
3251                                                 PF_POOLMASK(naddr,
3252                                                     &r->src.addr.p.dyn->
3253                                                     pfid_addr6,
3254                                                     &r->src.addr.p.dyn->
3255                                                     pfid_mask6,
3256                                                     daddr, AF_INET6);
3257                                                 break;
3258 #endif /* INET6 */
3259                                         }
3260                                 } else
3261                                         PF_POOLMASK(naddr,
3262                                             &r->src.addr.v.a.addr,
3263                                             &r->src.addr.v.a.mask, daddr,
3264                                             pd->af);
3265                                 break;
3266                         }
3267                         break;
3268                 case PF_RDR: {
3269                         m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3270                         if (pf_map_addr(pd->af, r, saddr, naddr, NULL, sn))
3271                                 return (NULL);
3272                         if ((r->rpool.opts & PF_POOL_TYPEMASK) ==
3273                             PF_POOL_BITMASK)
3274                                 PF_POOLMASK(naddr, naddr,
3275                                     &r->rpool.cur->addr.v.a.mask, daddr,
3276                                     pd->af);
3277
3278                         if (r->rpool.proxy_port[1]) {
3279                                 u_int32_t       tmp_nport;
3280
3281                                 tmp_nport = ((ntohs(dport) -
3282                                     ntohs(r->dst.port[0])) %
3283                                     (r->rpool.proxy_port[1] -
3284                                     r->rpool.proxy_port[0] + 1)) +
3285                                     r->rpool.proxy_port[0];
3286
3287                                 /* wrap around if necessary */
3288                                 if (tmp_nport > 65535)
3289                                         tmp_nport -= 65535;
3290                                 *nport = htons((u_int16_t)tmp_nport);
3291                         } else if (r->rpool.proxy_port[0]) {
3292                                 *nport = htons(r->rpool.proxy_port[0]);
3293                         }
3294                         pd->not_cpu_localized = 1;
3295                         break;
3296                 }
3297                 default:
3298                         return (NULL);
3299                 }
3300         }
3301
3302         return (r);
3303 }
3304
3305 struct netmsg_hashlookup {
3306         struct netmsg_base      base;
3307         struct inpcb            **nm_pinp;
3308         struct inpcbinfo        *nm_pcbinfo;
3309         struct pf_addr          *nm_saddr;
3310         struct pf_addr          *nm_daddr;
3311         uint16_t                nm_sport;
3312         uint16_t                nm_dport;
3313         sa_family_t             nm_af;
3314 };
3315
3316 #ifdef PF_SOCKET_LOOKUP_DOMSG
3317 static void
3318 in_pcblookup_hash_handler(netmsg_t msg)
3319 {
3320         struct netmsg_hashlookup *rmsg = (struct netmsg_hashlookup *)msg;
3321
3322         if (rmsg->nm_af == AF_INET)
3323                 *rmsg->nm_pinp = in_pcblookup_hash(rmsg->nm_pcbinfo,
3324                     rmsg->nm_saddr->v4, rmsg->nm_sport, rmsg->nm_daddr->v4,
3325                     rmsg->nm_dport, INPLOOKUP_WILDCARD, NULL);
3326 #ifdef INET6
3327         else
3328                 *rmsg->nm_pinp = in6_pcblookup_hash(rmsg->nm_pcbinfo,
3329                     &rmsg->nm_saddr->v6, rmsg->nm_sport, &rmsg->nm_daddr->v6,
3330                     rmsg->nm_dport, INPLOOKUP_WILDCARD, NULL);
3331 #endif /* INET6 */
3332         lwkt_replymsg(&rmsg->base.lmsg, 0);
3333 }
3334 #endif  /* PF_SOCKET_LOOKUP_DOMSG */
3335
3336 int
3337 pf_socket_lookup(int direction, struct pf_pdesc *pd)
3338 {
3339         struct pf_addr          *saddr, *daddr;
3340         u_int16_t                sport, dport;
3341         struct inpcbinfo        *pi;
3342         struct inpcb            *inp;
3343         struct netmsg_hashlookup *msg = NULL;
3344 #ifdef PF_SOCKET_LOOKUP_DOMSG
3345         struct netmsg_hashlookup msg0;
3346 #endif
3347         int                      pi_cpu = 0;
3348
3349         if (pd == NULL)
3350                 return (-1);
3351         pd->lookup.uid = UID_MAX;
3352         pd->lookup.gid = GID_MAX;
3353         pd->lookup.pid = NO_PID;
3354         if (direction == PF_IN) {
3355                 saddr = pd->src;
3356                 daddr = pd->dst;
3357         } else {
3358                 saddr = pd->dst;
3359                 daddr = pd->src;
3360         }
3361         switch (pd->proto) {
3362         case IPPROTO_TCP:
3363                 if (pd->hdr.tcp == NULL)
3364                         return (-1);
3365                 sport = pd->hdr.tcp->th_sport;
3366                 dport = pd->hdr.tcp->th_dport;
3367
3368                 pi_cpu = tcp_addrcpu(saddr->v4.s_addr, sport, daddr->v4.s_addr, dport);
3369                 pi = &tcbinfo[pi_cpu];
3370                 /*
3371                  * Our netstack runs lockless on MP systems
3372                  * (only for TCP connections at the moment).
3373                  *
3374                  * As we are not allowed to read another CPU's tcbinfo,
3375                  * we have to ask that CPU via remote call to search the
3376                  * table for us.
3377                  *
3378                  * Prepare a msg iff data belongs to another CPU.
3379                  */
3380                 if (pi_cpu != mycpu->gd_cpuid) {
3381 #ifdef PF_SOCKET_LOOKUP_DOMSG
3382                         /*
3383                          * NOTE:
3384                          *
3385                          * Following lwkt_domsg() is dangerous and could
3386                          * lockup the network system, e.g.
3387                          *
3388                          * On 2 CPU system:
3389                          * netisr0 domsg to netisr1 (due to lookup)
3390                          * netisr1 domsg to netisr0 (due to lookup)
3391                          *
3392                          * We simply return -1 here, since we are probably
3393                          * called before NAT, so the TCP packet should
3394                          * already be on the correct CPU.
3395                          */
3396                         msg = &msg0;
3397                         netmsg_init(&msg->base, NULL, &curthread->td_msgport,
3398                                     0, in_pcblookup_hash_handler);
3399                         msg->nm_pinp = &inp;
3400                         msg->nm_pcbinfo = pi;
3401                         msg->nm_saddr = saddr;
3402                         msg->nm_sport = sport;
3403                         msg->nm_daddr = daddr;
3404                         msg->nm_dport = dport;
3405                         msg->nm_af = pd->af;
3406 #else   /* !PF_SOCKET_LOOKUP_DOMSG */
3407                         kprintf("pf_socket_lookup: tcp packet not on the "
3408                                 "correct cpu %d, cur cpu %d\n",
3409                                 pi_cpu, mycpuid);
3410                         print_backtrace(-1);
3411                         return -1;
3412 #endif  /* PF_SOCKET_LOOKUP_DOMSG */
3413                 }
3414                 break;
3415         case IPPROTO_UDP:
3416                 if (pd->hdr.udp == NULL)
3417                         return (-1);
3418                 sport = pd->hdr.udp->uh_sport;
3419                 dport = pd->hdr.udp->uh_dport;
3420                 pi = &udbinfo[mycpuid];
3421                 break;
3422         default:
3423                 return (-1);
3424         }
3425         if (direction != PF_IN) {
3426                 u_int16_t       p;
3427
3428                 p = sport;
3429                 sport = dport;
3430                 dport = p;
3431         }
3432         switch (pd->af) {
3433 #ifdef INET6
3434         case AF_INET6:
3435                 /*
3436                  * Query other CPU, second part
3437                  *
3438                  * msg only gets initialized when:
3439                  * 1) packet is TCP
3440                  * 2) the info belongs to another CPU
3441                  *
3442                  * Use some switch/case magic to avoid code duplication.
3443                  */
3444                 if (msg == NULL) {
3445                         inp = in6_pcblookup_hash(pi, &saddr->v6, sport,
3446                             &daddr->v6, dport, INPLOOKUP_WILDCARD, NULL);
3447
3448                         if (inp == NULL)
3449                                 return (-1);
3450                         break;
3451                 }
3452                 /* FALLTHROUGH if SMP and on other CPU */
3453 #endif /* INET6 */
3454         case AF_INET:
3455                 if (msg != NULL) {
3456                         lwkt_domsg(netisr_cpuport(pi_cpu),
3457                                      &msg->base.lmsg, 0);
3458                 } else
3459                 {
3460                         inp = in_pcblookup_hash(pi, saddr->v4, sport, daddr->v4,
3461                             dport, INPLOOKUP_WILDCARD, NULL);
3462                 }
3463                 if (inp == NULL)
3464                         return (-1);
3465                 break;
3466
3467         default:
3468                 return (-1);
3469         }
3470         pd->lookup.uid = inp->inp_socket->so_cred->cr_uid;
3471         pd->lookup.gid = inp->inp_socket->so_cred->cr_groups[0];
3472         return (1);
3473 }
3474
3475 u_int8_t
3476 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3477 {
3478         int              hlen;
3479         u_int8_t         hdr[60];
3480         u_int8_t        *opt, optlen;
3481         u_int8_t         wscale = 0;
3482
3483         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
3484         if (hlen <= sizeof(struct tcphdr))
3485                 return (0);
3486         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3487                 return (0);
3488         opt = hdr + sizeof(struct tcphdr);
3489         hlen -= sizeof(struct tcphdr);
3490         while (hlen >= 3) {
3491                 switch (*opt) {
3492                 case TCPOPT_EOL:
3493                 case TCPOPT_NOP:
3494                         ++opt;
3495                         --hlen;
3496                         break;
3497                 case TCPOPT_WINDOW:
3498                         wscale = opt[2];
3499                         if (wscale > TCP_MAX_WINSHIFT)
3500                                 wscale = TCP_MAX_WINSHIFT;
3501                         wscale |= PF_WSCALE_FLAG;
3502                         /* FALLTHROUGH */
3503                 default:
3504                         optlen = opt[1];
3505                         if (optlen < 2)
3506                                 optlen = 2;
3507                         hlen -= optlen;
3508                         opt += optlen;
3509                         break;
3510                 }
3511         }
3512         return (wscale);
3513 }
3514
3515 u_int16_t
3516 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3517 {
3518         int              hlen;
3519         u_int8_t         hdr[60];
3520         u_int8_t        *opt, optlen;
3521         u_int16_t        mss = tcp_mssdflt;
3522
3523         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
3524         if (hlen <= sizeof(struct tcphdr))
3525                 return (0);
3526         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3527                 return (0);
3528         opt = hdr + sizeof(struct tcphdr);
3529         hlen -= sizeof(struct tcphdr);
3530         while (hlen >= TCPOLEN_MAXSEG) {
3531                 switch (*opt) {
3532                 case TCPOPT_EOL:
3533                 case TCPOPT_NOP:
3534                         ++opt;
3535                         --hlen;
3536                         break;
3537                 case TCPOPT_MAXSEG:
3538                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3539                         /* FALLTHROUGH */
3540                 default:
3541                         optlen = opt[1];
3542                         if (optlen < 2)
3543                                 optlen = 2;
3544                         hlen -= optlen;
3545                         opt += optlen;
3546                         break;
3547                 }
3548         }
3549         return (mss);
3550 }
3551
3552 u_int16_t
3553 pf_calc_mss(struct pf_addr *addr, sa_family_t af, u_int16_t offer)
3554 {
3555 #ifdef INET
3556         struct sockaddr_in      *dst;
3557         struct route             ro;
3558 #endif /* INET */
3559 #ifdef INET6
3560         struct sockaddr_in6     *dst6;
3561         struct route_in6         ro6;
3562 #endif /* INET6 */
3563         struct rtentry          *rt = NULL;
3564         int                      hlen = 0;
3565         u_int16_t                mss = tcp_mssdflt;
3566
3567         switch (af) {
3568 #ifdef INET
3569         case AF_INET:
3570                 hlen = sizeof(struct ip);
3571                 bzero(&ro, sizeof(ro));
3572                 dst = (struct sockaddr_in *)&ro.ro_dst;
3573                 dst->sin_family = AF_INET;
3574                 dst->sin_len = sizeof(*dst);
3575                 dst->sin_addr = addr->v4;
3576                 rtalloc_ign(&ro, (RTF_CLONING | RTF_PRCLONING));
3577                 rt = ro.ro_rt;
3578                 break;
3579 #endif /* INET */
3580 #ifdef INET6
3581         case AF_INET6:
3582                 hlen = sizeof(struct ip6_hdr);
3583                 bzero(&ro6, sizeof(ro6));
3584                 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
3585                 dst6->sin6_family = AF_INET6;
3586                 dst6->sin6_len = sizeof(*dst6);
3587                 dst6->sin6_addr = addr->v6;
3588                 rtalloc_ign((struct route *)&ro6, (RTF_CLONING | RTF_PRCLONING));
3589                 rt = ro6.ro_rt;
3590                 break;
3591 #endif /* INET6 */
3592         }
3593
3594         if (rt && rt->rt_ifp) {
3595                 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
3596                 mss = max(tcp_mssdflt, mss);
3597                 RTFREE(rt);
3598         }
3599         mss = min(mss, offer);
3600         mss = max(mss, 64);             /* sanity - at least max opt space */
3601         return (mss);
3602 }
3603
3604 void
3605 pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr)
3606 {
3607         struct pf_rule *r = s->rule.ptr;
3608
3609         s->rt_kif = NULL;
3610         if (!r->rt || r->rt == PF_FASTROUTE)
3611                 return;
3612         switch (s->key[PF_SK_WIRE]->af) {
3613 #ifdef INET
3614         case AF_INET:
3615                 pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL,
3616                     &s->nat_src_node);
3617                 s->rt_kif = r->rpool.cur->kif;
3618                 break;
3619 #endif /* INET */
3620 #ifdef INET6
3621         case AF_INET6:
3622                 pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL,
3623                     &s->nat_src_node);
3624                 s->rt_kif = r->rpool.cur->kif;
3625                 break;
3626 #endif /* INET6 */
3627         }
3628 }
3629
3630 u_int32_t
3631 pf_tcp_iss(struct pf_pdesc *pd)
3632 {
3633         MD5_CTX ctx;
3634         u_int32_t digest[4];
3635
3636         if (pf_tcp_secret_init == 0) {
3637                 lwkt_gettoken(&pf_gtoken);
3638                 if (pf_tcp_secret_init == 0) {
3639                         karc4rand(pf_tcp_secret, sizeof(pf_tcp_secret));
3640                         MD5Init(&pf_tcp_secret_ctx);
3641                         MD5Update(&pf_tcp_secret_ctx, pf_tcp_secret,
3642                             sizeof(pf_tcp_secret));
3643                         pf_tcp_secret_init = 1;
3644                 }
3645                 lwkt_reltoken(&pf_gtoken);
3646         }
3647         ctx = pf_tcp_secret_ctx;
3648
3649         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3650         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3651         if (pd->af == AF_INET6) {
3652                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3653                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3654         } else {
3655                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3656                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3657         }
3658         MD5Final((u_char *)digest, &ctx);
3659         pf_tcp_iss_off += 4096;
3660
3661         return (digest[0] + pd->hdr.tcp->th_seq + pf_tcp_iss_off);
3662 }
3663
3664 int
3665 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3666     struct pfi_kif *kif, struct mbuf *m, int off, void *h,
3667     struct pf_pdesc *pd, struct pf_rule **am, struct pf_ruleset **rsm,
3668     struct ifqueue *ifq, struct inpcb *inp)
3669 {
3670         struct pf_rule          *nr = NULL;
3671         struct pf_addr          *saddr = pd->src, *daddr = pd->dst;
3672         sa_family_t              af = pd->af;
3673         struct pf_rule          *r, *a = NULL;
3674         struct pf_ruleset       *ruleset = NULL;
3675         struct pf_src_node      *nsn = NULL;
3676         struct tcphdr           *th = pd->hdr.tcp;
3677         struct pf_state_key     *skw = NULL, *sks = NULL;
3678         struct pf_state_key     *sk = NULL, *nk = NULL;
3679         u_short                  reason;
3680         int                      rewrite = 0, hdrlen = 0;
3681         int                      tag = -1, rtableid = -1;
3682         int                      asd = 0;
3683         int                      match = 0;
3684         int                      state_icmp = 0;
3685         u_int16_t                sport = 0, dport = 0;
3686         u_int16_t                bproto_sum = 0, bip_sum = 0;
3687         u_int8_t                 icmptype = 0, icmpcode = 0;
3688
3689
3690         if (direction == PF_IN && pf_check_congestion(ifq)) {
3691                 REASON_SET(&reason, PFRES_CONGEST);
3692                 return (PF_DROP);
3693         }
3694
3695         if (inp != NULL)
3696                 pd->lookup.done = pf_socket_lookup(direction, pd);
3697         else if (debug_pfugidhack) {
3698                 DPFPRINTF(PF_DEBUG_MISC, ("pf: unlocked lookup\n"));
3699                 pd->lookup.done = pf_socket_lookup(direction, pd);
3700         }
3701
3702         switch (pd->proto) {
3703         case IPPROTO_TCP:
3704                 sport = th->th_sport;
3705                 dport = th->th_dport;
3706                 hdrlen = sizeof(*th);
3707                 break;
3708         case IPPROTO_UDP:
3709                 sport = pd->hdr.udp->uh_sport;
3710                 dport = pd->hdr.udp->uh_dport;
3711                 hdrlen = sizeof(*pd->hdr.udp);
3712                 break;
3713 #ifdef INET
3714         case IPPROTO_ICMP:
3715                 if (pd->af != AF_INET)
3716                         break;
3717                 sport = dport = pd->hdr.icmp->icmp_id;
3718                 hdrlen = sizeof(*pd->hdr.icmp);
3719                 icmptype = pd->hdr.icmp->icmp_type;
3720                 icmpcode = pd->hdr.icmp->icmp_code;
3721
3722                 if (icmptype == ICMP_UNREACH ||
3723                     icmptype == ICMP_SOURCEQUENCH ||
3724                     icmptype == ICMP_REDIRECT ||
3725                     icmptype == ICMP_TIMXCEED ||
3726                     icmptype == ICMP_PARAMPROB)
3727                         state_icmp++;
3728                 break;
3729 #endif /* INET */
3730 #ifdef INET6
3731         case IPPROTO_ICMPV6:
3732                 if (af != AF_INET6)
3733                         break;
3734                 sport = dport = pd->hdr.icmp6->icmp6_id;
3735                 hdrlen = sizeof(*pd->hdr.icmp6);
3736                 icmptype = pd->hdr.icmp6->icmp6_type;
3737                 icmpcode = pd->hdr.icmp6->icmp6_code;
3738
3739                 if (icmptype == ICMP6_DST_UNREACH ||
3740                     icmptype == ICMP6_PACKET_TOO_BIG ||
3741                     icmptype == ICMP6_TIME_EXCEEDED ||
3742                     icmptype == ICMP6_PARAM_PROB)
3743                         state_icmp++;
3744                 break;
3745 #endif /* INET6 */
3746         default:
3747                 sport = dport = hdrlen = 0;
3748                 break;
3749         }
3750
3751         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3752
3753         /* check packet for BINAT/NAT/RDR */
3754         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn,
3755             &skw, &sks, &sk, &nk, saddr, daddr, sport, dport)) != NULL) {
3756                 if (nk == NULL || sk == NULL) {
3757                         REASON_SET(&reason, PFRES_MEMORY);
3758                         goto cleanup;
3759                 }
3760
3761                 if (pd->ip_sum)
3762                         bip_sum = *pd->ip_sum;
3763
3764                 m->m_flags &= ~M_HASH;
3765                 switch (pd->proto) {
3766                 case IPPROTO_TCP:
3767                         bproto_sum = th->th_sum;
3768                         pd->proto_sum = &th->th_sum;
3769
3770                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3771                             nk->port[pd->sidx] != sport) {
3772                                 pf_change_ap(saddr, &th->th_sport, pd->ip_sum,
3773                                     &th->th_sum, &nk->addr[pd->sidx],
3774                                     nk->port[pd->sidx], 0, af);
3775                                 pd->sport = &th->th_sport;
3776                                 sport = th->th_sport;
3777                         }
3778
3779                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3780                             nk->port[pd->didx] != dport) {
3781                                 pf_change_ap(daddr, &th->th_dport, pd->ip_sum,
3782                                     &th->th_sum, &nk->addr[pd->didx],
3783                                     nk->port[pd->didx], 0, af);
3784                                 dport = th->th_dport;
3785                                 pd->dport = &th->th_dport;
3786                         }
3787                         rewrite++;
3788                         break;
3789                 case IPPROTO_UDP:
3790                         bproto_sum = pd->hdr.udp->uh_sum;
3791                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3792
3793                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3794                             nk->port[pd->sidx] != sport) {
3795                                 pf_change_ap(saddr, &pd->hdr.udp->uh_sport,
3796                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3797                                     &nk->addr[pd->sidx],
3798                                     nk->port[pd->sidx], 1, af);
3799                                 sport = pd->hdr.udp->uh_sport;
3800                                 pd->sport = &pd->hdr.udp->uh_sport;
3801                         }
3802
3803                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3804                             nk->port[pd->didx] != dport) {
3805                                 pf_change_ap(daddr, &pd->hdr.udp->uh_dport,
3806                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3807                                     &nk->addr[pd->didx],
3808                                     nk->port[pd->didx], 1, af);
3809                                 dport = pd->hdr.udp->uh_dport;
3810                                 pd->dport = &pd->hdr.udp->uh_dport;
3811                         }
3812                         rewrite++;
3813                         break;
3814 #ifdef INET
3815                 case IPPROTO_ICMP:
3816                         nk->port[0] = nk->port[1];
3817                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3818                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3819                                     nk->addr[pd->sidx].v4.s_addr, 0);
3820
3821                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3822                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3823                                     nk->addr[pd->didx].v4.s_addr, 0);
3824
3825                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3826                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3827                                     pd->hdr.icmp->icmp_cksum, sport,
3828                                     nk->port[1], 0);
3829                                 pd->hdr.icmp->icmp_id = nk->port[1];
3830                                 pd->sport = &pd->hdr.icmp->icmp_id;
3831                         }
3832                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3833                         break;
3834 #endif /* INET */
3835 #ifdef INET6
3836                 case IPPROTO_ICMPV6:
3837                         nk->port[0] = nk->port[1];
3838                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3839                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3840                                     &nk->addr[pd->sidx], 0);
3841
3842                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3843                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3844                                     &nk->addr[pd->didx], 0);
3845                         rewrite++;
3846                         break;
3847 #endif /* INET */
3848                 default:
3849                         switch (af) {
3850 #ifdef INET
3851                         case AF_INET:
3852                                 if (PF_ANEQ(saddr,
3853                                     &nk->addr[pd->sidx], AF_INET))
3854                                         pf_change_a(&saddr->v4.s_addr,
3855                                             pd->ip_sum,
3856                                             nk->addr[pd->sidx].v4.s_addr, 0);
3857
3858                                 if (PF_ANEQ(daddr,
3859                                     &nk->addr[pd->didx], AF_INET))
3860                                         pf_change_a(&daddr->v4.s_addr,
3861                                             pd->ip_sum,
3862                                             nk->addr[pd->didx].v4.s_addr, 0);
3863                                 break;
3864 #endif /* INET */
3865 #ifdef INET6
3866                         case AF_INET6:
3867                                 if (PF_ANEQ(saddr,
3868                                     &nk->addr[pd->sidx], AF_INET6))
3869                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3870
3871                                 if (PF_ANEQ(daddr,
3872                                     &nk->addr[pd->didx], AF_INET6))
3873                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3874                                 break;
3875 #endif /* INET */
3876                         }
3877                         break;
3878                 }
3879                 if (nr->natpass)
3880                         r = NULL;
3881                 pd->nat_rule = nr;
3882         }
3883
3884         while (r != NULL) {
3885                 r->evaluations++;
3886                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3887                         r = r->skip[PF_SKIP_IFP].ptr;
3888                 else if (r->direction && r->direction != direction)
3889                         r = r->skip[PF_SKIP_DIR].ptr;
3890                 else if (r->af && r->af != af)
3891                         r = r->skip[PF_SKIP_AF].ptr;
3892                 else if (r->proto && r->proto != pd->proto)
3893                         r = r->skip[PF_SKIP_PROTO].ptr;
3894                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3895                     r->src.neg, kif))
3896                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3897                 /* tcp/udp only. port_op always 0 in other cases */
3898                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3899                     r->src.port[0], r->src.port[1], sport))
3900                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3901                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3902                     r->dst.neg, NULL))
3903                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3904                 /* tcp/udp only. port_op always 0 in other cases */
3905                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3906                     r->dst.port[0], r->dst.port[1], dport))
3907                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3908                 /* icmp only. type always 0 in other cases */
3909                 else if (r->type && r->type != icmptype + 1)
3910                         r = TAILQ_NEXT(r, entries);
3911                 /* icmp only. type always 0 in other cases */
3912                 else if (r->code && r->code != icmpcode + 1)
3913                         r = TAILQ_NEXT(r, entries);
3914                 else if (r->tos && !(r->tos == pd->tos))
3915                         r = TAILQ_NEXT(r, entries);
3916                 else if (r->rule_flag & PFRULE_FRAGMENT)
3917                         r = TAILQ_NEXT(r, entries);
3918                 else if (pd->proto == IPPROTO_TCP &&
3919                     (r->flagset & th->th_flags) != r->flags)
3920                         r = TAILQ_NEXT(r, entries);
3921                 /* tcp/udp only. uid.op always 0 in other cases */
3922                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3923                     pf_socket_lookup(direction, pd), 1)) &&
3924                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3925                     pd->lookup.uid))
3926                         r = TAILQ_NEXT(r, entries);
3927                 /* tcp/udp only. gid.op always 0 in other cases */
3928                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3929                     pf_socket_lookup(direction, pd), 1)) &&
3930                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3931                     pd->lookup.gid))
3932                         r = TAILQ_NEXT(r, entries);
3933                 else if (r->prob &&
3934                   r->prob <= karc4random())
3935                         r = TAILQ_NEXT(r, entries);
3936                 else if (r->match_tag && !pf_match_tag(m, r, &tag))
3937                         r = TAILQ_NEXT(r, entries);
3938                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3939                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3940                     pf_osfp_fingerprint(pd, m, off, th),
3941                     r->os_fingerprint)))
3942                         r = TAILQ_NEXT(r, entries);
3943                 else {
3944                         if (r->tag)
3945                                 tag = r->tag;
3946                         if (r->rtableid >= 0)
3947                                 rtableid = r->rtableid;
3948                         if (r->anchor == NULL) {
3949                                 match = 1;
3950                                 *rm = r;
3951                                 *am = a;
3952                                 *rsm = ruleset;
3953                                 if ((*rm)->quick)
3954                                         break;
3955                                 r = TAILQ_NEXT(r, entries);
3956                         } else
3957                                 pf_step_into_anchor(&asd, &ruleset,
3958                                     PF_RULESET_FILTER, &r, &a, &match);
3959                 }
3960                 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
3961                     PF_RULESET_FILTER, &r, &a, &match))
3962                         break;
3963         }
3964         r = *rm;
3965         a = *am;
3966         ruleset = *rsm;
3967
3968         REASON_SET(&reason, PFRES_MATCH);
3969
3970         if (r->log || (nr != NULL && nr->log)) {
3971                 if (rewrite)
3972                         m_copyback(m, off, hdrlen, pd->hdr.any);
3973                 PFLOG_PACKET(kif, h, m, af, direction, reason, r->log ? r : nr,
3974                     a, ruleset, pd);
3975         }
3976
3977         if ((r->action == PF_DROP) &&
3978             ((r->rule_flag & PFRULE_RETURNRST) ||
3979             (r->rule_flag & PFRULE_RETURNICMP) ||
3980             (r->rule_flag & PFRULE_RETURN))) {
3981                 /* undo NAT changes, if they have taken place */
3982                 if (nr != NULL) {
3983                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3984                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3985                         if (pd->sport)
3986                                 *pd->sport = sk->port[pd->sidx];
3987                         if (pd->dport)
3988                                 *pd->dport = sk->port[pd->didx];
3989                         if (pd->proto_sum)
3990                                 *pd->proto_sum = bproto_sum;
3991                         if (pd->ip_sum)
3992                                 *pd->ip_sum = bip_sum;
3993                         m_copyback(m, off, hdrlen, pd->hdr.any);
3994                 }
3995                 if (pd->proto == IPPROTO_TCP &&
3996                     ((r->rule_flag & PFRULE_RETURNRST) ||
3997                     (r->rule_flag & PFRULE_RETURN)) &&
3998                     !(th->th_flags & TH_RST)) {
3999                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
4000                         int              len = 0;
4001                         struct ip       *h4;
4002 #ifdef INET6
4003                         struct ip6_hdr  *h6;
4004 #endif
4005                         switch (af) {
4006                         case AF_INET:
4007                                 h4 = mtod(m, struct ip *);
4008                                 len = h4->ip_len - off;
4009                                 break;
4010 #ifdef INET6
4011                         case AF_INET6:
4012                                 h6 = mtod(m, struct ip6_hdr *);
4013                                 len = h6->ip6_plen - (off - sizeof(*h6));
4014                                 break;
4015 #endif
4016                         }
4017
4018                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
4019                                 REASON_SET(&reason, PFRES_PROTCKSUM);
4020                         else {
4021                                 if (th->th_flags & TH_SYN)
4022                                         ack++;
4023                                 if (th->th_flags & TH_FIN)
4024                                         ack++;
4025                                 pf_send_tcp(r, af, pd->dst,
4026                                     pd->src, th->th_dport, th->th_sport,
4027                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
4028                                     r->return_ttl, 1, 0, pd->eh, kif->pfik_ifp);
4029                         }
4030                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
4031                     r->return_icmp)
4032                         pf_send_icmp(m, r->return_icmp >> 8,
4033                             r->return_icmp & 255, af, r);
4034                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
4035                     r->return_icmp6)
4036                         pf_send_icmp(m, r->return_icmp6 >> 8,
4037                             r->return_icmp6 & 255, af, r);
4038         }
4039
4040         if (r->action == PF_DROP)
4041                 goto cleanup;
4042
4043         if (pf_tag_packet(m, tag, rtableid)) {
4044                 REASON_SET(&reason, PFRES_MEMORY);
4045                 goto cleanup;
4046         }
4047
4048         if (!state_icmp && (r->keep_state || nr != NULL ||
4049             (pd->flags & PFDESC_TCP_NORM))) {
4050                 int action;
4051                 action = pf_create_state(r, nr, a, pd, nsn, skw, sks, nk, sk, m,
4052                     off, sport, dport, &rewrite, kif, sm, tag, bproto_sum,
4053                     bip_sum, hdrlen);
4054                 if (action != PF_PASS)
4055                         return (action);
4056         }
4057
4058         /* copy back packet headers if we performed NAT operations */
4059         if (rewrite)
4060                 m_copyback(m, off, hdrlen, pd->hdr.any);
4061
4062         return (PF_PASS);
4063
4064 cleanup:
4065         if (sk != NULL)
4066                 kfree(sk, M_PFSTATEKEYPL);
4067         if (nk != NULL)
4068                 kfree(nk, M_PFSTATEKEYPL);
4069         return (PF_DROP);
4070 }
4071
4072 static __inline int
4073 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
4074     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *skw,
4075     struct pf_state_key *sks, struct pf_state_key *nk, struct pf_state_key *sk,
4076     struct mbuf *m, int off, u_int16_t sport, u_int16_t dport, int *rewrite,
4077     struct pfi_kif *kif, struct pf_state **sm, int tag, u_int16_t bproto_sum,
4078     u_int16_t bip_sum, int hdrlen)
4079 {
4080         struct pf_state         *s = NULL;
4081         struct pf_src_node      *sn = NULL;
4082         struct tcphdr           *th = pd->hdr.tcp;
4083         u_int16_t                mss = tcp_mssdflt;
4084         u_short                  reason;
4085         int cpu = mycpu->gd_cpuid;
4086
4087         /* check maximums */
4088         if (r->max_states && (r->states_cur >= r->max_states)) {
4089                 PF_INC_LCOUNTER(LCNT_STATES);
4090                 REASON_SET(&reason, PFRES_MAXSTATES);
4091                 return (PF_DROP);
4092         }
4093         /* src node for filter rule */
4094         if ((r->rule_flag & PFRULE_SRCTRACK ||
4095             r->rpool.opts & PF_POOL_STICKYADDR) &&
4096             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
4097                 REASON_SET(&reason, PFRES_SRCLIMIT);
4098                 goto csfailed;
4099         }
4100         /* src node for translation rule */
4101         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
4102             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
4103                 REASON_SET(&reason, PFRES_SRCLIMIT);
4104                 goto csfailed;
4105         }
4106         s = kmalloc(sizeof(struct pf_state), M_PFSTATEPL, M_NOWAIT|M_ZERO);
4107         if (s == NULL) {
4108                 REASON_SET(&reason, PFRES_MEMORY);
4109                 goto csfailed;
4110         }
4111         lockinit(&s->lk, "pfstlk", 0, 0);
4112         s->id = 0; /* XXX Do we really need that? not in OpenBSD */
4113         s->creatorid = 0;
4114         s->rule.ptr = r;
4115         s->nat_rule.ptr = nr;
4116         s->anchor.ptr = a;
4117         s->state_flags = PFSTATE_CREATEINPROG;
4118         STATE_INC_COUNTERS(s);
4119         if (r->allow_opts)
4120                 s->state_flags |= PFSTATE_ALLOWOPTS;
4121         if (r->rule_flag & PFRULE_STATESLOPPY)
4122                 s->state_flags |= PFSTATE_SLOPPY;
4123         if (pd->not_cpu_localized)
4124                 s->state_flags |= PFSTATE_STACK_GLOBAL;
4125
4126         s->log = r->log & PF_LOG_ALL;
4127         if (nr != NULL)
4128                 s->log |= nr->log & PF_LOG_ALL;
4129         switch (pd->proto) {
4130         case IPPROTO_TCP:
4131                 s->src.seqlo = ntohl(th->th_seq);
4132                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
4133                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
4134                     r->keep_state == PF_STATE_MODULATE) {
4135                         /* Generate sequence number modulator */
4136                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
4137                             0)
4138                                 s->src.seqdiff = 1;
4139                         pf_change_a(&th->th_seq, &th->th_sum,
4140                             htonl(s->src.seqlo + s->src.seqdiff), 0);
4141                         *rewrite = 1;
4142                 } else
4143                         s->src.seqdiff = 0;
4144                 if (th->th_flags & TH_SYN) {
4145                         s->src.seqhi++;
4146                         s->src.wscale = pf_get_wscale(m, off,
4147                             th->th_off, pd->af);
4148                 }
4149                 s->src.max_win = MAX(ntohs(th->th_win), 1);
4150                 if (s->src.wscale & PF_WSCALE_MASK) {
4151                         /* Remove scale factor from initial window */
4152                         int win = s->src.max_win;
4153                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
4154                         s->src.max_win = (win - 1) >>
4155                             (s->src.wscale & PF_WSCALE_MASK);
4156                 }
4157                 if (th->th_flags & TH_FIN)
4158                         s->src.seqhi++;
4159                 s->dst.seqhi = 1;
4160                 s->dst.max_win = 1;
4161                 s->src.state = TCPS_SYN_SENT;
4162                 s->dst.state = TCPS_CLOSED;
4163                 s->timeout = PFTM_TCP_FIRST_PACKET;
4164                 break;
4165         case IPPROTO_UDP:
4166                 s->src.state = PFUDPS_SINGLE;
4167                 s->dst.state = PFUDPS_NO_TRAFFIC;
4168                 s->timeout = PFTM_UDP_FIRST_PACKET;
4169                 break;
4170         case IPPROTO_ICMP:
4171 #ifdef INET6
4172         case IPPROTO_ICMPV6:
4173 #endif
4174                 s->timeout = PFTM_ICMP_FIRST_PACKET;
4175                 break;
4176         default:
4177                 s->src.state = PFOTHERS_SINGLE;
4178                 s->dst.state = PFOTHERS_NO_TRAFFIC;
4179                 s->timeout = PFTM_OTHER_FIRST_PACKET;
4180         }
4181
4182         s->creation = time_second;
4183         s->expire = time_second;
4184
4185         if (sn != NULL) {
4186                 s->src_node = sn;
4187                 s->src_node->states++;
4188         }
4189         if (nsn != NULL) {
4190                 /* XXX We only modify one side for now. */
4191                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
4192                 s->nat_src_node = nsn;
4193                 s->nat_src_node->states++;
4194         }
4195         if (pd->proto == IPPROTO_TCP) {
4196                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
4197                     off, pd, th, &s->src, &s->dst)) {
4198                         REASON_SET(&reason, PFRES_MEMORY);
4199                         pf_src_tree_remove_state(s);
4200                         STATE_DEC_COUNTERS(s);
4201                         kfree(s, M_PFSTATEPL);
4202                         return (PF_DROP);
4203                 }
4204                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
4205                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
4206                     &s->src, &s->dst, rewrite)) {
4207                         /* This really shouldn't happen!!! */
4208                         DPFPRINTF(PF_DEBUG_URGENT,
4209                             ("pf_normalize_tcp_stateful failed on first pkt"));
4210                         pf_normalize_tcp_cleanup(s);
4211                         pf_src_tree_remove_state(s);
4212                         STATE_DEC_COUNTERS(s);
4213                         kfree(s, M_PFSTATEPL);
4214                         return (PF_DROP);
4215                 }
4216         }
4217         s->direction = pd->dir;
4218
4219         if (sk == NULL && pf_state_key_setup(pd, nr, &skw, &sks, &sk, &nk,
4220                                              pd->src, pd->dst, sport, dport)) {
4221                 REASON_SET(&reason, PFRES_MEMORY);
4222                 goto csfailed;
4223         }
4224
4225         if (pf_state_insert(BOUND_IFACE(r, kif), skw, sks, s)) {
4226                 if (pd->proto == IPPROTO_TCP)
4227                         pf_normalize_tcp_cleanup(s);
4228                 REASON_SET(&reason, PFRES_STATEINS);
4229                 pf_src_tree_remove_state(s);
4230                 STATE_DEC_COUNTERS(s);
4231                 kfree(s, M_PFSTATEPL);
4232                 return (PF_DROP);
4233         } else
4234                 *sm = s;
4235
4236         pf_set_rt_ifp(s, pd->src);      /* needs s->state_key set */
4237         if (tag > 0) {
4238                 pf_tag_ref(tag);
4239                 s->tag = tag;
4240         }
4241         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
4242             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
4243                 s->src.state = PF_TCPS_PROXY_SRC;
4244                 /* undo NAT changes, if they have taken place */
4245                 if (nr != NULL) {
4246                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
4247                         if (pd->dir == PF_OUT)
4248                                 skt = s->key[PF_SK_STACK];
4249                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
4250                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
4251                         if (pd->sport)
4252                                 *pd->sport = skt->port[pd->sidx];
4253                         if (pd->dport)
4254                                 *pd->dport = skt->port[pd->didx];
4255                         if (pd->proto_sum)
4256                                 *pd->proto_sum = bproto_sum;
4257                         if (pd->ip_sum)
4258                                 *pd->ip_sum = bip_sum;
4259                         m->m_flags &= ~M_HASH;
4260                         m_copyback(m, off, hdrlen, pd->hdr.any);
4261                 }
4262                 s->src.seqhi = htonl(karc4random());
4263                 /* Find mss option */
4264                 mss = pf_get_mss(m, off, th->th_off, pd->af);
4265                 mss = pf_calc_mss(pd->src, pd->af, mss);
4266                 mss = pf_calc_mss(pd->dst, pd->af, mss);
4267                 s->src.mss = mss;
4268                 s->state_flags &= ~PFSTATE_CREATEINPROG;
4269                 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport,
4270                             th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
4271                             TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL, NULL);
4272                 REASON_SET(&reason, PFRES_SYNPROXY);
4273                 return (PF_SYNPROXY_DROP);
4274         }
4275
4276         s->state_flags &= ~PFSTATE_CREATEINPROG;
4277         return (PF_PASS);
4278
4279 csfailed:
4280         if (sk != NULL)
4281                 kfree(sk, M_PFSTATEKEYPL);
4282         if (nk != NULL)
4283                 kfree(nk, M_PFSTATEKEYPL);
4284
4285         if (sn != NULL && sn->states == 0 && sn->expire == 0) {
4286                 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], sn);
4287                 PF_INC_SCOUNTER(SCNT_SRC_NODE_REMOVALS);
4288                 atomic_add_int(&pf_status.src_nodes, -1);
4289                 kfree(sn, M_PFSRCTREEPL);
4290         }
4291         if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0) {
4292                 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], nsn);
4293                 PF_INC_SCOUNTER(SCNT_SRC_NODE_REMOVALS);
4294                 atomic_add_int(&pf_status.src_nodes, -1);
4295                 kfree(nsn, M_PFSRCTREEPL);
4296         }
4297         if (s) {
4298                 pf_src_tree_remove_state(s);
4299                 STATE_DEC_COUNTERS(s);
4300                 kfree(s, M_PFSTATEPL);
4301         }
4302
4303         return (PF_DROP);
4304 }
4305
4306 int
4307 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
4308     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
4309     struct pf_ruleset **rsm)
4310 {
4311         struct pf_rule          *r, *a = NULL;
4312         struct pf_ruleset       *ruleset = NULL;
4313         sa_family_t              af = pd->af;
4314         u_short                  reason;
4315         int                      tag = -1;
4316         int                      asd = 0;
4317         int                      match = 0;
4318
4319         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
4320         while (r != NULL) {
4321                 r->evaluations++;
4322                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
4323                         r = r->skip[PF_SKIP_IFP].ptr;
4324                 else if (r->direction && r->direction != direction)
4325                         r = r->skip[PF_SKIP_DIR].ptr;
4326                 else if (r->af && r->af != af)
4327                         r = r->skip[PF_SKIP_AF].ptr;
4328                 else if (r->proto && r->proto != pd->proto)
4329                         r = r->skip[PF_SKIP_PROTO].ptr;
4330                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
4331                     r->src.neg, kif))
4332                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
4333                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
4334                     r->dst.neg, NULL))
4335                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
4336                 else if (r->tos && !(r->tos == pd->tos))
4337                         r = TAILQ_NEXT(r, entries);
4338                 else if (r->os_fingerprint != PF_OSFP_ANY)
4339                         r = TAILQ_NEXT(r, entries);
4340                 else if (pd->proto == IPPROTO_UDP &&
4341                     (r->src.port_op || r->dst.port_op))
4342                         r = TAILQ_NEXT(r, entries);
4343                 else if (pd->proto == IPPROTO_TCP &&
4344                     (r->src.port_op || r->dst.port_op || r->flagset))
4345                         r = TAILQ_NEXT(r, entries);
4346                 else if ((pd->proto == IPPROTO_ICMP ||
4347                     pd->proto == IPPROTO_ICMPV6) &&
4348                     (r->type || r->code))
4349                         r = TAILQ_NEXT(r, entries);
4350                 else if (r->prob && r->prob <= karc4random())
4351                         r = TAILQ_NEXT(r, entries);
4352                 else if (r->match_tag && !pf_match_tag(m, r, &tag))
4353                         r = TAILQ_NEXT(r, entries);
4354                 else {
4355                         if (r->anchor == NULL) {
4356                                 match = 1;
4357                                 *rm = r;
4358                                 *am = a;
4359                                 *rsm = ruleset;
4360                                 if ((*rm)->quick)
4361                                         break;
4362                                 r = TAILQ_NEXT(r, entries);
4363                         } else
4364                                 pf_step_into_anchor(&asd, &ruleset,
4365                                     PF_RULESET_FILTER, &r, &a, &match);
4366                 }
4367                 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
4368                     PF_RULESET_FILTER, &r, &a, &match))
4369                         break;
4370         }
4371         r = *rm;
4372         a = *am;
4373         ruleset = *rsm;
4374
4375         REASON_SET(&reason, PFRES_MATCH);
4376
4377         if (r->log)
4378                 PFLOG_PACKET(kif, h, m, af, direction, reason, r, a, ruleset,
4379                     pd);
4380
4381         if (r->action != PF_PASS)
4382                 return (PF_DROP);
4383
4384         if (pf_tag_packet(m, tag, -1)) {
4385                 REASON_SET(&reason, PFRES_MEMORY);
4386                 return (PF_DROP);
4387         }
4388
4389         return (PF_PASS);
4390 }
4391
4392 /*
4393  * Called with state locked
4394  */
4395 int
4396 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
4397         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
4398         struct pf_pdesc *pd, u_short *reason, int *copyback)
4399 {
4400         struct tcphdr           *th = pd->hdr.tcp;
4401         u_int16_t                win = ntohs(th->th_win);
4402         u_int32_t                ack, end, seq, orig_seq;
4403         u_int8_t                 sws, dws;
4404         int                      ackskew;
4405
4406         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
4407                 sws = src->wscale & PF_WSCALE_MASK;
4408                 dws = dst->wscale & PF_WSCALE_MASK;
4409         } else {
4410                 sws = dws = 0;
4411         }
4412
4413         /*
4414          * Sequence tracking algorithm from Guido van Rooij's paper:
4415          *   http://www.madison-gurkha.com/publications/tcp_filtering/
4416          *      tcp_filtering.ps
4417          */
4418
4419         orig_seq = seq = ntohl(th->th_seq);
4420         if (src->seqlo == 0) {
4421                 /* First packet from this end. Set its state */
4422
4423                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
4424                     src->scrub == NULL) {
4425                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
4426                                 REASON_SET(reason, PFRES_MEMORY);
4427                                 return (PF_DROP);
4428                         }
4429                 }
4430
4431                 /* Deferred generation of sequence number modulator */
4432                 if (dst->seqdiff && !src->seqdiff) {
4433                         /* use random iss for the TCP server */
4434                         while ((src->seqdiff = karc4random() - seq) == 0)
4435                                 ;
4436                         ack = ntohl(th->th_ack) - dst->seqdiff;
4437                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
4438                             src->seqdiff), 0);
4439                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
4440                         *copyback = 1;
4441                 } else {
4442                         ack = ntohl(th->th_ack);
4443                 }
4444
4445                 end = seq + pd->p_len;
4446                 if (th->th_flags & TH_SYN) {
4447                         end++;
4448                         (*state)->sync_flags |= PFSTATE_GOT_SYN2;
4449                         if (dst->wscale & PF_WSCALE_FLAG) {
4450                                 src->wscale = pf_get_wscale(m, off, th->th_off,
4451                                     pd->af);
4452                                 if (src->wscale & PF_WSCALE_FLAG) {
4453                                         /* Remove scale factor from initial
4454                                          * window */
4455                                         sws = src->wscale & PF_WSCALE_MASK;
4456                                         win = ((u_int32_t)win + (1 << sws) - 1)
4457                                             >> sws;
4458                                         dws = dst->wscale & PF_WSCALE_MASK;
4459                                 } else {
4460                                         /* fixup other window */
4461                                         dst->max_win <<= dst->wscale &
4462                                             PF_WSCALE_MASK;
4463                                         /* in case of a retrans SYN|ACK */
4464                                         dst->wscale = 0;
4465                                 }
4466                         }
4467                 }
4468                 if (th->th_flags & TH_FIN)
4469                         end++;
4470
4471                 src->seqlo = seq;
4472                 if (src->state < TCPS_SYN_SENT)
4473                         src->state = TCPS_SYN_SENT;
4474
4475                 /*
4476                  * May need to slide the window (seqhi may have been set by
4477                  * the crappy stack check or if we picked up the connection
4478                  * after establishment)
4479                  */
4480                 if (src->seqhi == 1 ||
4481                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
4482                         src->seqhi = end + MAX(1, dst->max_win << dws);
4483                 if (win > src->max_win)
4484                         src->max_win = win;
4485
4486         } else {
4487                 ack = ntohl(th->th_ack) - dst->seqdiff;
4488                 if (src->seqdiff) {
4489                         /* Modulate sequence numbers */
4490                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
4491                             src->seqdiff), 0);
4492                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
4493                         *copyback = 1;
4494                 }
4495                 end = seq + pd->p_len;
4496                 if (th->th_flags & TH_SYN)
4497                         end++;
4498                 if (th->th_flags & TH_FIN)
4499                         end++;
4500         }
4501
4502         if ((th->th_flags & TH_ACK) == 0) {
4503                 /* Let it pass through the ack skew check */
4504                 ack = dst->seqlo;
4505         } else if ((ack == 0 &&
4506             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4507             /* broken tcp stacks do not set ack */
4508             (dst->state < TCPS_SYN_SENT)) {
4509                 /*
4510                  * Many stacks (ours included) will set the ACK number in an
4511                  * FIN|ACK if the SYN times out -- no sequence to ACK.
4512                  */
4513                 ack = dst->seqlo;
4514         }
4515
4516         if (seq == end) {
4517                 /* Ease sequencing restrictions on no data packets */
4518                 seq = src->seqlo;
4519                 end = seq;
4520         }
4521
4522         ackskew = dst->seqlo - ack;
4523
4524
4525         /*
4526          * Need to demodulate the sequence numbers in any TCP SACK options
4527          * (Selective ACK). We could optionally validate the SACK values
4528          * against the current ACK window, either forwards or backwards, but
4529          * I'm not confident that SACK has been implemented properly
4530          * everywhere. It wouldn't surprise me if several stacks accidently
4531          * SACK too far backwards of previously ACKed data. There really aren't
4532          * any security implications of bad SACKing unless the target stack
4533          * doesn't validate the option length correctly. Someone trying to
4534          * spoof into a TCP connection won't bother blindly sending SACK
4535          * options anyway.
4536          */
4537         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4538                 if (pf_modulate_sack(m, off, pd, th, dst))
4539                         *copyback = 1;
4540         }
4541
4542
4543 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
4544         if (SEQ_GEQ(src->seqhi, end) &&
4545             /* Last octet inside other's window space */
4546             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4547             /* Retrans: not more than one window back */
4548             (ackskew >= -MAXACKWINDOW) &&
4549             /* Acking not more than one reassembled fragment backwards */
4550             (ackskew <= (MAXACKWINDOW << sws)) &&
4551             /* Acking not more than one window forward */
4552             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4553             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4554             (pd->flags & PFDESC_IP_REAS) == 0)) {
4555             /* Require an exact/+1 sequence match on resets when possible */
4556
4557                 if (dst->scrub || src->scrub) {
4558                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4559                             *state, src, dst, copyback))
4560                                 return (PF_DROP);
4561                 }
4562
4563                 /* update max window */
4564                 if (src->max_win < win)
4565                         src->max_win = win;
4566                 /* synchronize sequencing */
4567                 if (SEQ_GT(end, src->seqlo))
4568                         src->seqlo = end;
4569                 /* slide the window of what the other end can send */
4570                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4571                         dst->seqhi = ack + MAX((win << sws), 1);
4572
4573
4574                 /* update states */
4575                 if (th->th_flags & TH_SYN)
4576                         if (src->state < TCPS_SYN_SENT)
4577                                 src->state = TCPS_SYN_SENT;
4578                 if (th->th_flags & TH_FIN)
4579                         if (src->state < TCPS_CLOSING)
4580                                 src->state = TCPS_CLOSING;
4581                 if (th->th_flags & TH_ACK) {
4582                         if (dst->state == TCPS_SYN_SENT) {
4583                                 dst->state = TCPS_ESTABLISHED;
4584                                 if (src->state == TCPS_ESTABLISHED &&
4585                                     (*state)->src_node != NULL &&
4586                                     pf_src_connlimit(*state)) {
4587                                         REASON_SET(reason, PFRES_SRCLIMIT);
4588                                         return (PF_DROP);
4589                                 }
4590                         } else if (dst->state == TCPS_CLOSING)
4591                                 dst->state = TCPS_FIN_WAIT_2;
4592                 }
4593                 if (th->th_flags & TH_RST)
4594                         src->state = dst->state = TCPS_TIME_WAIT;
4595
4596                 /* update expire time */
4597                 (*state)->expire = time_second;
4598                 if (src->state >= TCPS_FIN_WAIT_2 &&
4599                     dst->state >= TCPS_FIN_WAIT_2)
4600                         (*state)->timeout = PFTM_TCP_CLOSED;
4601                 else if (src->state >= TCPS_CLOSING &&
4602                     dst->state >= TCPS_CLOSING)
4603                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
4604                 else if (src->state < TCPS_ESTABLISHED ||
4605                     dst->state < TCPS_ESTABLISHED)
4606                         (*state)->timeout = PFTM_TCP_OPENING;
4607                 else if (src->state >= TCPS_CLOSING ||
4608                     dst->state >= TCPS_CLOSING)
4609                         (*state)->timeout = PFTM_TCP_CLOSING;
4610                 else
4611                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4612
4613                 /* Fall through to PASS packet */
4614
4615         } else if ((dst->state < TCPS_SYN_SENT ||
4616                 dst->state >= TCPS_FIN_WAIT_2 ||
4617                 src->state >= TCPS_FIN_WAIT_2) &&
4618             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4619             /* Within a window forward of the originating packet */
4620             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4621             /* Within a window backward of the originating packet */
4622
4623                 /*
4624                  * This currently handles three situations:
4625                  *  1) Stupid stacks will shotgun SYNs before their peer
4626                  *     replies.
4627                  *  2) When PF catches an already established stream (the
4628                  *     firewall rebooted, the state table was flushed, routes
4629                  *     changed...)
4630                  *  3) Packets get funky immediately after the connection
4631                  *     closes (this should catch Solaris spurious ACK|FINs
4632                  *     that web servers like to spew after a close)
4633                  *
4634                  * This must be a little more careful than the above code
4635                  * since packet floods will also be caught here. We don't
4636                  * update the TTL here to mitigate the damage of a packet
4637                  * flood and so the same code can handle awkward establishment
4638                  * and a loosened connection close.
4639                  * In the establishment case, a correct peer response will
4640                  * validate the connection, go through the normal state code
4641                  * and keep updating the state TTL.
4642                  */
4643
4644                 if (pf_status.debug >= PF_DEBUG_MISC) {
4645                         kprintf("pf: loose state match: ");
4646                         pf_print_state(*state);
4647                         pf_print_flags(th->th_flags);
4648                         kprintf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4649                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, pd->p_len,
4650                             ackskew, (unsigned long long)(*state)->packets[0],
4651                             (unsigned long long)(*state)->packets[1],
4652                             pd->dir == PF_IN ? "in" : "out",
4653                             pd->dir == (*state)->direction ? "fwd" : "rev");
4654                 }
4655
4656                 if (dst->scrub || src->scrub) {
4657                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4658                             *state, src, dst, copyback))
4659                                 return (PF_DROP);
4660                 }
4661
4662                 /* update max window */
4663                 if (src->max_win < win)
4664                         src->max_win = win;
4665                 /* synchronize sequencing */
4666                 if (SEQ_GT(end, src->seqlo))
4667                         src->seqlo = end;
4668                 /* slide the window of what the other end can send */
4669                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4670                         dst->seqhi = ack + MAX((win << sws), 1);
4671
4672                 /*
4673                  * Cannot set dst->seqhi here since this could be a shotgunned
4674                  * SYN and not an already established connection.
4675                  */
4676
4677                 if (th->th_flags & TH_FIN)
4678                         if (src->state < TCPS_CLOSING)
4679                                 src->state = TCPS_CLOSING;
4680                 if (th->th_flags & TH_RST)
4681                         src->state = dst->state = TCPS_TIME_WAIT;
4682
4683                 /* Fall through to PASS packet */
4684
4685         } else if ((*state)->pickup_mode == PF_PICKUPS_HASHONLY ||
4686                     ((*state)->pickup_mode == PF_PICKUPS_ENABLED &&
4687                      ((*state)->sync_flags & PFSTATE_GOT_SYN_MASK) !=
4688                       PFSTATE_GOT_SYN_MASK)) {
4689                 /*
4690                  * If pickup mode is hash only, do not fail on sequence checks.
4691                  *
4692                  * If pickup mode is enabled and we did not see the SYN in
4693                  * both direction, do not fail on sequence checks because
4694                  * we do not have complete information on window scale.
4695                  *
4696                  * Adjust expiration and fall through to PASS packet.
4697                  * XXX Add a FIN check to reduce timeout?
4698                  */
4699                 (*state)->expire = time_second;
4700         } else  {
4701                 /*
4702                  * Failure processing
4703                  */
4704                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4705                     (*state)->src.state == TCPS_SYN_SENT) {
4706                         /* Send RST for state mismatches during handshake */
4707                         if (!(th->th_flags & TH_RST))
4708                                 pf_send_tcp((*state)->rule.ptr, pd->af,
4709                                     pd->dst, pd->src, th->th_dport,
4710                                     th->th_sport, ntohl(th->th_ack), 0,
4711                                     TH_RST, 0, 0,
4712                                     (*state)->rule.ptr->return_ttl, 1, 0,
4713                                     pd->eh, kif->pfik_ifp);
4714                         src->seqlo = 0;
4715                         src->seqhi = 1;
4716                         src->max_win = 1;
4717                 } else if (pf_status.debug >= PF_DEBUG_MISC) {
4718                         kprintf("pf: BAD state: ");
4719                         pf_print_state(*state);
4720                         pf_print_flags(th->th_flags);
4721                         kprintf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4722                             "pkts=%llu:%llu dir=%s,%s\n",
4723                             seq, orig_seq, ack, pd->p_len, ackskew,
4724                             (unsigned long long)(*state)->packets[0],
4725                                 (unsigned long long)(*state)->packets[1],
4726                             pd->dir == PF_IN ? "in" : "out",
4727                             pd->dir == (*state)->direction ? "fwd" : "rev");
4728                         kprintf("pf: State failure on: %c %c %c %c | %c %c\n",
4729                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4730                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4731                             ' ': '2',
4732                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4733                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4734                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4735                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4736                 }
4737                 REASON_SET(reason, PFRES_BADSTATE);
4738                 return (PF_DROP);
4739         }
4740
4741         return (PF_PASS);
4742 }
4743
4744 /*
4745  * Called with state locked
4746  */
4747 int
4748 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4749         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4750 {
4751         struct tcphdr           *th = pd->hdr.tcp;
4752
4753         if (th->th_flags & TH_SYN)
4754                 if (src->state < TCPS_SYN_SENT)
4755                         src->state = TCPS_SYN_SENT;
4756         if (th->th_flags & TH_FIN)
4757                 if (src->state < TCPS_CLOSING)
4758                         src->state = TCPS_CLOSING;
4759         if (th->th_flags & TH_ACK) {
4760                 if (dst->state == TCPS_SYN_SENT) {
4761                         dst->state = TCPS_ESTABLISHED;
4762                         if (src->state == TCPS_ESTABLISHED &&
4763                             (*state)->src_node != NULL &&
4764                             pf_src_connlimit(*state)) {
4765                                 REASON_SET(reason, PFRES_SRCLIMIT);
4766                                 return (PF_DROP);
4767                         }
4768                 } else if (dst->state == TCPS_CLOSING) {
4769                         dst->state = TCPS_FIN_WAIT_2;
4770                 } else if (src->state == TCPS_SYN_SENT &&
4771                     dst->state < TCPS_SYN_SENT) {
4772                         /*
4773                          * Handle a special sloppy case where we only see one
4774                          * half of the connection. If there is a ACK after
4775                          * the initial SYN without ever seeing a packet from
4776                          * the destination, set the connection to established.
4777                          */
4778                         dst->state = src->state = TCPS_ESTABLISHED;
4779                         if ((*state)->src_node != NULL &&
4780                             pf_src_connlimit(*state)) {
4781                                 REASON_SET(reason, PFRES_SRCLIMIT);
4782                                 return (PF_DROP);
4783                         }
4784                 } else if (src->state == TCPS_CLOSING &&
4785                     dst->state == TCPS_ESTABLISHED &&
4786                     dst->seqlo == 0) {
4787                         /*
4788                          * Handle the closing of half connections where we
4789                          * don't see the full bidirectional FIN/ACK+ACK
4790                          * handshake.
4791                          */
4792                         dst->state = TCPS_CLOSING;
4793                 }
4794         }
4795         if (th->th_flags & TH_RST)
4796                 src->state = dst->state = TCPS_TIME_WAIT;
4797
4798         /* update expire time */
4799         (*state)->expire = time_second;
4800         if (src->state >= TCPS_FIN_WAIT_2 &&
4801             dst->state >= TCPS_FIN_WAIT_2)
4802                 (*state)->timeout = PFTM_TCP_CLOSED;
4803         else if (src->state >= TCPS_CLOSING &&
4804             dst->state >= TCPS_CLOSING)
4805                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4806         else if (src->state < TCPS_ESTABLISHED ||
4807             dst->state < TCPS_ESTABLISHED)
4808                 (*state)->timeout = PFTM_TCP_OPENING;
4809         else if (src->state >= TCPS_CLOSING ||
4810             dst->state >= TCPS_CLOSING)
4811                 (*state)->timeout = PFTM_TCP_CLOSING;
4812         else
4813                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4814
4815         return (PF_PASS);
4816 }
4817
4818 /*
4819  * Test TCP connection state.  Caller must hold the state locked.
4820  */
4821 int
4822 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4823                   struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4824                   u_short *reason)
4825 {
4826         struct pf_state_key_cmp  key;
4827         struct tcphdr           *th = pd->hdr.tcp;
4828         int                      copyback = 0;
4829         int                      error;
4830         struct pf_state_peer    *src, *dst;
4831         struct pf_state_key     *sk;
4832
4833         bzero(&key, sizeof(key));
4834         key.af = pd->af;
4835         key.proto = IPPROTO_TCP;
4836         if (direction == PF_IN) {       /* wire side, straight */
4837                 PF_ACPY(&key.addr[0], pd->src, key.af);
4838                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4839                 key.port[0] = th->th_sport;
4840                 key.port[1] = th->th_dport;
4841                 if (pf_status.debug >= PF_DEBUG_MISC) {
4842                         kprintf("test-tcp IN (%08x:%d) -> (%08x:%d)\n",
4843                                 ntohl(key.addr[0].addr32[0]),
4844                                 ntohs(key.port[0]),
4845                                 ntohl(key.addr[1].addr32[0]),
4846                                 ntohs(key.port[1]));
4847                 }
4848         } else {                        /* stack side, reverse */
4849                 PF_ACPY(&key.addr[1], pd->src, key.af);
4850                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4851                 key.port[1] = th->th_sport;
4852                 key.port[0] = th->th_dport;
4853                 if (pf_status.debug >= PF_DEBUG_MISC) {
4854                         kprintf("test-tcp OUT (%08x:%d) <- (%08x:%d)\n",
4855                                 ntohl(key.addr[0].addr32[0]),
4856                                 ntohs(key.port[0]),
4857                                 ntohl(key.addr[1].addr32[0]),
4858                                 ntohs(key.port[1]));
4859                 }
4860         }
4861
4862         STATE_LOOKUP(kif, &key, direction, *state, m);
4863         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
4864
4865         if (direction == (*state)->direction) {
4866                 src = &(*state)->src;
4867                 dst = &(*state)->dst;
4868         } else {
4869                 src = &(*state)->dst;
4870                 dst = &(*state)->src;
4871         }
4872
4873         sk = (*state)->key[pd->didx];
4874
4875         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4876                 if (direction != (*state)->direction) {
4877                         REASON_SET(reason, PFRES_SYNPROXY);
4878                         FAIL (PF_SYNPROXY_DROP);
4879                 }
4880                 if (th->th_flags & TH_SYN) {
4881                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4882                                 REASON_SET(reason, PFRES_SYNPROXY);
4883                                 FAIL (PF_DROP);
4884                         }
4885                         pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
4886                             pd->src, th->th_dport, th->th_sport,
4887                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4888                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1,
4889                             0, NULL, NULL);
4890                         REASON_SET(reason, PFRES_SYNPROXY);
4891                         FAIL (PF_SYNPROXY_DROP);
4892                 } else if (!(th->th_flags & TH_ACK) ||
4893                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4894                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4895                         REASON_SET(reason, PFRES_SYNPROXY);
4896                         FAIL (PF_DROP);
4897                 } else if ((*state)->src_node != NULL &&
4898                     pf_src_connlimit(*state)) {
4899                         REASON_SET(reason, PFRES_SRCLIMIT);
4900                         FAIL (PF_DROP);
4901                 } else
4902                         (*state)->src.state = PF_TCPS_PROXY_DST;
4903         }
4904         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4905                 if (direction == (*state)->direction) {
4906                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4907                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4908                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4909                                 REASON_SET(reason, PFRES_SYNPROXY);
4910                                 FAIL (PF_DROP);
4911                         }
4912                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4913                         if ((*state)->dst.seqhi == 1)
4914                                 (*state)->dst.seqhi = htonl(karc4random());
4915                         pf_send_tcp((*state)->rule.ptr, pd->af,
4916                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4917                             sk->port[pd->sidx], sk->port[pd->didx],
4918                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4919                             (*state)->src.mss, 0, 0, (*state)->tag, NULL, NULL);
4920                         REASON_SET(reason, PFRES_SYNPROXY);
4921                         FAIL (PF_SYNPROXY_DROP);
4922                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4923                     (TH_SYN|TH_ACK)) ||
4924                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4925                         REASON_SET(reason, PFRES_SYNPROXY);
4926                         FAIL (PF_DROP);
4927                 } else {
4928                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4929                         (*state)->dst.seqlo = ntohl(th->th_seq);
4930                         pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
4931                             pd->src, th->th_dport, th->th_sport,
4932                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4933                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4934                             (*state)->tag, NULL, NULL);
4935                         pf_send_tcp((*state)->rule.ptr, pd->af,
4936                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4937                             sk->port[pd->sidx], sk->port[pd->didx],
4938                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4939                             TH_ACK, (*state)->dst.max_win, 0, 0, 1,
4940                             0, NULL, NULL);
4941                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4942                             (*state)->src.seqlo;
4943                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4944                             (*state)->dst.seqlo;
4945                         (*state)->src.seqhi = (*state)->src.seqlo +
4946                             (*state)->dst.max_win;
4947                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4948                             (*state)->src.max_win;
4949                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4950                         (*state)->src.state = (*state)->dst.state =
4951                             TCPS_ESTABLISHED;
4952                         REASON_SET(reason, PFRES_SYNPROXY);
4953                         FAIL (PF_SYNPROXY_DROP);
4954                 }
4955         }
4956
4957         /*
4958          * Check for connection (addr+port pair) reuse.  We can't actually
4959          * unlink the state if we don't own it.
4960          */
4961         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4962             dst->state >= TCPS_FIN_WAIT_2 &&
4963             src->state >= TCPS_FIN_WAIT_2) {
4964                 if (pf_status.debug >= PF_DEBUG_MISC) {
4965                         kprintf("pf: state reuse ");
4966                         pf_print_state(*state);
4967                         pf_print_flags(th->th_flags);
4968                         kprintf("\n");
4969                 }
4970                 /* XXX make sure it's the same direction ?? */
4971                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4972                 if ((*state)->cpuid == mycpu->gd_cpuid) {
4973                         pf_unlink_state(*state);
4974                         *state = NULL;
4975                 } else {
4976                         (*state)->timeout = PFTM_PURGE;
4977                 }
4978                 FAIL (PF_DROP);
4979         }
4980
4981         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4982                 if (pf_tcp_track_sloppy(src, dst, state, pd,
4983                                         reason) == PF_DROP) {
4984                         FAIL (PF_DROP);
4985                 }
4986         } else {
4987                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd,
4988                                       reason, &copyback) == PF_DROP) {
4989                         FAIL (PF_DROP);
4990                 }
4991         }
4992
4993         /* translate source/destination address, if necessary */
4994         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4995                 struct pf_state_key *nk = (*state)->key[pd->didx];
4996
4997                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4998                     nk->port[pd->sidx] != th->th_sport)  {
4999                         /*
5000                          * The translated source address may be completely
5001                          * unrelated to the saved link header, make sure
5002                          * a bridge doesn't try to use it.
5003                          */
5004                         m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
5005                         pf_change_ap(pd->src, &th->th_sport, pd->ip_sum,
5006                             &th->th_sum, &nk->addr[pd->sidx],
5007                             nk->port[pd->sidx], 0, pd->af);
5008                 }
5009
5010                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5011                     nk->port[pd->didx] != th->th_dport) {
5012                         /*
5013                          * If we don't redispatch the packet will go into
5014                          * the protocol stack on the wrong cpu for the
5015                          * post-translated address.
5016                          */
5017                         pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum,
5018                             &th->th_sum, &nk->addr[pd->didx],
5019                             nk->port[pd->didx], 0, pd->af);
5020                 }
5021                 copyback = 1;
5022         }
5023
5024         /* Copyback sequence modulation or stateful scrub changes if needed */
5025         if (copyback) {
5026                 m->m_flags &= ~M_HASH;
5027                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
5028         }
5029
5030         pfsync_update_state(*state);
5031         error = PF_PASS;
5032 done:
5033         if (*state)
5034                 lockmgr(&(*state)->lk, LK_RELEASE);
5035         return (error);
5036 }
5037
5038 /*
5039  * Test UDP connection state.  Caller must hold the state locked.
5040  */
5041 int
5042 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
5043                   struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
5044 {
5045         struct pf_state_peer    *src, *dst;
5046         struct pf_state_key_cmp  key;
5047         struct udphdr           *uh = pd->hdr.udp;
5048
5049         bzero(&key, sizeof(key));
5050         key.af = pd->af;
5051         key.proto = IPPROTO_UDP;
5052         if (direction == PF_IN) {       /* wire side, straight */
5053                 PF_ACPY(&key.addr[0], pd->src, key.af);
5054                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5055                 key.port[0] = uh->uh_sport;
5056                 key.port[1] = uh->uh_dport;
5057         } else {                        /* stack side, reverse */
5058                 PF_ACPY(&key.addr[1], pd->src, key.af);
5059                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5060                 key.port[1] = uh->uh_sport;
5061                 key.port[0] = uh->uh_dport;
5062         }
5063
5064         STATE_LOOKUP(kif, &key, direction, *state, m);
5065         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5066
5067         if (direction == (*state)->direction) {
5068                 src = &(*state)->src;
5069                 dst = &(*state)->dst;
5070         } else {
5071                 src = &(*state)->dst;
5072                 dst = &(*state)->src;
5073         }
5074
5075         /* update states */
5076         if (src->state < PFUDPS_SINGLE)
5077                 src->state = PFUDPS_SINGLE;
5078         if (dst->state == PFUDPS_SINGLE)
5079                 dst->state = PFUDPS_MULTIPLE;
5080
5081         /* update expire time */
5082         (*state)->expire = time_second;
5083         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
5084                 (*state)->timeout = PFTM_UDP_MULTIPLE;
5085         else
5086                 (*state)->timeout = PFTM_UDP_SINGLE;
5087
5088         /* translate source/destination address, if necessary */
5089         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5090                 struct pf_state_key *nk = (*state)->key[pd->didx];
5091
5092                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5093                     nk->port[pd->sidx] != uh->uh_sport) {
5094                         /*
5095                          * The translated source address may be completely
5096                          * unrelated to the saved link header, make sure
5097                          * a bridge doesn't try to use it.
5098                          */
5099                         m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
5100                         m->m_flags &= ~M_HASH;
5101                         pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum,
5102                             &uh->uh_sum, &nk->addr[pd->sidx],
5103                             nk->port[pd->sidx], 1, pd->af);
5104                 }
5105
5106                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5107                     nk->port[pd->didx] != uh->uh_dport) {
5108                         /*
5109                          * If we don't redispatch the packet will go into
5110                          * the protocol stack on the wrong cpu for the
5111                          * post-translated address.
5112                          */
5113                         m->m_flags &= ~M_HASH;
5114                         pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum,
5115                             &uh->uh_sum, &nk->addr[pd->didx],
5116                             nk->port[pd->didx], 1, pd->af);
5117                 }
5118                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
5119         }
5120
5121         pfsync_update_state(*state);
5122         lockmgr(&(*state)->lk, LK_RELEASE);
5123         return (PF_PASS);
5124 }
5125
5126 /*
5127  * Test ICMP connection state.  Caller must hold the state locked.
5128  */
5129 int
5130 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
5131                    struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
5132                    u_short *reason)
5133 {
5134         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
5135         u_int16_t        icmpid = 0, *icmpsum = NULL;
5136         u_int8_t         icmptype = 0;
5137         int              state_icmp = 0;
5138         int              error;
5139         struct pf_state_key_cmp key;
5140
5141         bzero(&key, sizeof(key));
5142
5143         switch (pd->proto) {
5144 #ifdef INET
5145         case IPPROTO_ICMP:
5146                 icmptype = pd->hdr.icmp->icmp_type;
5147                 icmpid = pd->hdr.icmp->icmp_id;
5148                 icmpsum = &pd->hdr.icmp->icmp_cksum;
5149
5150                 if (icmptype == ICMP_UNREACH ||
5151                     icmptype == ICMP_SOURCEQUENCH ||
5152                     icmptype == ICMP_REDIRECT ||
5153                     icmptype == ICMP_TIMXCEED ||
5154                     icmptype == ICMP_PARAMPROB)
5155                         state_icmp++;
5156                 break;
5157 #endif /* INET */
5158 #ifdef INET6
5159         case IPPROTO_ICMPV6:
5160                 icmptype = pd->hdr.icmp6->icmp6_type;
5161                 icmpid = pd->hdr.icmp6->icmp6_id;
5162                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
5163
5164                 if (icmptype == ICMP6_DST_UNREACH ||
5165                     icmptype == ICMP6_PACKET_TOO_BIG ||
5166                     icmptype == ICMP6_TIME_EXCEEDED ||
5167                     icmptype == ICMP6_PARAM_PROB)
5168                         state_icmp++;
5169                 break;
5170 #endif /* INET6 */
5171         }
5172
5173         if (!state_icmp) {
5174
5175                 /*
5176                  * ICMP query/reply message not related to a TCP/UDP packet.
5177                  * Search for an ICMP state.
5178                  */
5179                 key.af = pd->af;
5180                 key.proto = pd->proto;
5181                 key.port[0] = key.port[1] = icmpid;
5182                 if (direction == PF_IN) {       /* wire side, straight */
5183                         PF_ACPY(&key.addr[0], pd->src, key.af);
5184                         PF_ACPY(&key.addr[1], pd->dst, key.af);
5185                 } else {                        /* stack side, reverse */
5186                         PF_ACPY(&key.addr[1], pd->src, key.af);
5187                         PF_ACPY(&key.addr[0], pd->dst, key.af);
5188                 }
5189
5190                 STATE_LOOKUP(kif, &key, direction, *state, m);
5191                 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5192
5193                 (*state)->expire = time_second;
5194                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
5195
5196                 /* translate source/destination address, if necessary */
5197                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5198                         struct pf_state_key *nk = (*state)->key[pd->didx];
5199
5200                         switch (pd->af) {
5201 #ifdef INET
5202                         case AF_INET:
5203                                 if (PF_ANEQ(pd->src,
5204                                     &nk->addr[pd->sidx], AF_INET))
5205                                         pf_change_a(&saddr->v4.s_addr,
5206                                             pd->ip_sum,
5207                                             nk->addr[pd->sidx].v4.s_addr, 0);
5208
5209                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
5210                                     AF_INET))
5211                                         pf_change_a(&daddr->v4.s_addr,
5212                                             pd->ip_sum,
5213                                             nk->addr[pd->didx].v4.s_addr, 0);
5214
5215                                 if (nk->port[0] !=
5216                                     pd->hdr.icmp->icmp_id) {
5217                                         pd->hdr.icmp->icmp_cksum =
5218                                             pf_cksum_fixup(
5219                                             pd->hdr.icmp->icmp_cksum, icmpid,
5220                                             nk->port[pd->sidx], 0);
5221                                         pd->hdr.icmp->icmp_id =
5222                                             nk->port[pd->sidx];
5223                                 }
5224
5225                                 m->m_flags &= ~M_HASH;
5226                                 m_copyback(m, off, ICMP_MINLEN,
5227                                     (caddr_t)pd->hdr.icmp);
5228                                 break;
5229 #endif /* INET */
5230 #ifdef INET6
5231                         case AF_INET6:
5232                                 if (PF_ANEQ(pd->src,
5233                                     &nk->addr[pd->sidx], AF_INET6))
5234                                         pf_change_a6(saddr,
5235                                             &pd->hdr.icmp6->icmp6_cksum,
5236                                             &nk->addr[pd->sidx], 0);
5237
5238                                 if (PF_ANEQ(pd->dst,
5239                                     &nk->addr[pd->didx], AF_INET6))
5240                                         pf_change_a6(daddr,
5241                                             &pd->hdr.icmp6->icmp6_cksum,
5242                                             &nk->addr[pd->didx], 0);
5243
5244                                 m->m_flags &= ~M_HASH;
5245                                 m_copyback(m, off,
5246                                         sizeof(struct icmp6_hdr),
5247                                         (caddr_t)pd->hdr.icmp6);
5248                                 break;
5249 #endif /* INET6 */
5250                         }
5251                 }
5252         } else {
5253                 /*
5254                  * ICMP error message in response to a TCP/UDP packet.
5255                  * Extract the inner TCP/UDP header and search for that state.
5256                  */
5257
5258                 struct pf_pdesc pd2;
5259 #ifdef INET
5260                 struct ip       h2;
5261 #endif /* INET */
5262 #ifdef INET6
5263                 struct ip6_hdr  h2_6;
5264                 int             terminal = 0;
5265 #endif /* INET6 */
5266                 int             ipoff2;
5267                 int             off2;
5268
5269                 pd2.not_cpu_localized = 1;
5270                 pd2.af = pd->af;
5271                 /* Payload packet is from the opposite direction. */
5272                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
5273                 pd2.didx = (direction == PF_IN) ? 0 : 1;
5274                 switch (pd->af) {
5275 #ifdef INET
5276                 case AF_INET:
5277                         /* offset of h2 in mbuf chain */
5278                         ipoff2 = off + ICMP_MINLEN;
5279
5280                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
5281                             NULL, reason, pd2.af)) {
5282                                 DPFPRINTF(PF_DEBUG_MISC,
5283                                     ("pf: ICMP error message too short "
5284                                     "(ip)\n"));
5285                                 FAIL (PF_DROP);
5286                         }
5287                         /*
5288                          * ICMP error messages don't refer to non-first
5289                          * fragments
5290                          */
5291                         if (h2.ip_off & htons(IP_OFFMASK)) {
5292                                 REASON_SET(reason, PFRES_FRAG);
5293                                 FAIL (PF_DROP);
5294                         }
5295
5296                         /* offset of protocol header that follows h2 */
5297                         off2 = ipoff2 + (h2.ip_hl << 2);
5298
5299                         pd2.proto = h2.ip_p;
5300                         pd2.src = (struct pf_addr *)&h2.ip_src;
5301                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
5302                         pd2.ip_sum = &h2.ip_sum;
5303                         break;
5304 #endif /* INET */
5305 #ifdef INET6
5306                 case AF_INET6:
5307                         ipoff2 = off + sizeof(struct icmp6_hdr);
5308
5309                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
5310                             NULL, reason, pd2.af)) {
5311                                 DPFPRINTF(PF_DEBUG_MISC,
5312                                     ("pf: ICMP error message too short "
5313                                     "(ip6)\n"));
5314                                 FAIL (PF_DROP);
5315                         }
5316                         pd2.proto = h2_6.ip6_nxt;
5317                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
5318                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
5319                         pd2.ip_sum = NULL;
5320                         off2 = ipoff2 + sizeof(h2_6);
5321                         do {
5322                                 switch (pd2.proto) {
5323                                 case IPPROTO_FRAGMENT:
5324                                         /*
5325                                          * ICMPv6 error messages for
5326                                          * non-first fragments
5327                                          */
5328                                         REASON_SET(reason, PFRES_FRAG);
5329                                         FAIL (PF_DROP);
5330                                 case IPPROTO_AH:
5331                                 case IPPROTO_HOPOPTS:
5332                                 case IPPROTO_ROUTING:
5333                                 case IPPROTO_DSTOPTS: {
5334                                         /* get next header and header length */
5335                                         struct ip6_ext opt6;
5336
5337                                         if (!pf_pull_hdr(m, off2, &opt6,
5338                                             sizeof(opt6), NULL, reason,
5339                                             pd2.af)) {
5340                                                 DPFPRINTF(PF_DEBUG_MISC,
5341                                                     ("pf: ICMPv6 short opt\n"));
5342                                                 FAIL (PF_DROP);
5343                                         }
5344                                         if (pd2.proto == IPPROTO_AH)
5345                                                 off2 += (opt6.ip6e_len + 2) * 4;
5346                                         else
5347                                                 off2 += (opt6.ip6e_len + 1) * 8;
5348                                         pd2.proto = opt6.ip6e_nxt;
5349                                         /* goto the next header */
5350                                         break;
5351                                 }
5352                                 default:
5353                                         terminal++;
5354                                         break;
5355                                 }
5356                         } while (!terminal);
5357                         break;
5358 #endif /* INET6 */
5359                 default:
5360                         DPFPRINTF(PF_DEBUG_MISC,
5361                             ("pf: ICMP AF %d unknown (ip6)\n", pd->af));
5362                         FAIL (PF_DROP);
5363                         break;
5364                 }
5365
5366                 switch (pd2.proto) {
5367                 case IPPROTO_TCP: {
5368                         struct tcphdr            th;
5369                         u_int32_t                seq;
5370                         struct pf_state_peer    *src, *dst;
5371                         u_int8_t                 dws;
5372                         int                      copyback = 0;
5373
5374                         /*
5375                          * Only the first 8 bytes of the TCP header can be
5376                          * expected. Don't access any TCP header fields after
5377                          * th_seq, an ackskew test is not possible.
5378                          */
5379                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
5380                             pd2.af)) {
5381                                 DPFPRINTF(PF_DEBUG_MISC,
5382                                     ("pf: ICMP error message too short "
5383                                     "(tcp)\n"));
5384                                 FAIL (PF_DROP);
5385                         }
5386
5387                         key.af = pd2.af;
5388                         key.proto = IPPROTO_TCP;
5389                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5390                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5391                         key.port[pd2.sidx] = th.th_sport;
5392                         key.port[pd2.didx] = th.th_dport;
5393
5394                         STATE_LOOKUP(kif, &key, direction, *state, m);
5395                         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5396
5397                         if (direction == (*state)->direction) {
5398                                 src = &(*state)->dst;
5399                                 dst = &(*state)->src;
5400                         } else {
5401                                 src = &(*state)->src;
5402                                 dst = &(*state)->dst;
5403                         }
5404
5405                         if (src->wscale && dst->wscale)
5406                                 dws = dst->wscale & PF_WSCALE_MASK;
5407                         else
5408                                 dws = 0;
5409
5410                         /* Demodulate sequence number */
5411                         seq = ntohl(th.th_seq) - src->seqdiff;
5412                         if (src->seqdiff) {
5413                                 pf_change_a(&th.th_seq, icmpsum,
5414                                     htonl(seq), 0);
5415                                 copyback = 1;
5416                         }
5417
5418                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
5419                             (!SEQ_GEQ(src->seqhi, seq) ||
5420                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
5421                                 if (pf_status.debug >= PF_DEBUG_MISC) {
5422                                         kprintf("pf: BAD ICMP %d:%d ",
5423                                             icmptype, pd->hdr.icmp->icmp_code);
5424                                         pf_print_host(pd->src, 0, pd->af);
5425                                         kprintf(" -> ");
5426                                         pf_print_host(pd->dst, 0, pd->af);
5427                                         kprintf(" state: ");
5428                                         pf_print_state(*state);
5429                                         kprintf(" seq=%u\n", seq);
5430                                 }
5431                                 REASON_SET(reason, PFRES_BADSTATE);
5432                                 FAIL (PF_DROP);
5433                         } else {
5434                                 if (pf_status.debug >= PF_DEBUG_MISC) {
5435                                         kprintf("pf: OK ICMP %d:%d ",
5436                                             icmptype, pd->hdr.icmp->icmp_code);
5437                                         pf_print_host(pd->src, 0, pd->af);
5438                                         kprintf(" -> ");
5439                                         pf_print_host(pd->dst, 0, pd->af);
5440                                         kprintf(" state: ");
5441                                         pf_print_state(*state);
5442                                         kprintf(" seq=%u\n", seq);
5443                                 }
5444                         }
5445
5446                         /* translate source/destination address, if necessary */
5447                         if ((*state)->key[PF_SK_WIRE] !=
5448                             (*state)->key[PF_SK_STACK]) {
5449                                 struct pf_state_key *nk =
5450                                     (*state)->key[pd->didx];
5451
5452                                 if (PF_ANEQ(pd2.src,
5453                                     &nk->addr[pd2.sidx], pd2.af) ||
5454                                     nk->port[pd2.sidx] != th.th_sport)
5455                                         pf_change_icmp(pd2.src, &th.th_sport,
5456                                             daddr, &nk->addr[pd2.sidx],
5457                                             nk->port[pd2.sidx], NULL,
5458                                             pd2.ip_sum, icmpsum,
5459                                             pd->ip_sum, 0, pd2.af);
5460
5461                                 if (PF_ANEQ(pd2.dst,
5462                                     &nk->addr[pd2.didx], pd2.af) ||
5463                                     nk->port[pd2.didx] != th.th_dport)
5464                                         pf_change_icmp(pd2.dst, &th.th_dport,
5465                                             NULL, /* XXX Inbound NAT? */
5466                                             &nk->addr[pd2.didx],
5467                                             nk->port[pd2.didx], NULL,
5468                                             pd2.ip_sum, icmpsum,
5469                                             pd->ip_sum, 0, pd2.af);
5470                                 copyback = 1;
5471                         }
5472
5473                         if (copyback) {
5474                                 switch (pd2.af) {
5475 #ifdef INET
5476                                 case AF_INET:
5477                                         m_copyback(m, off, ICMP_MINLEN,
5478                                             (caddr_t)pd->hdr.icmp);
5479                                         m_copyback(m, ipoff2, sizeof(h2),
5480                                             (caddr_t)&h2);
5481                                         break;
5482 #endif /* INET */
5483 #ifdef INET6
5484                                 case AF_INET6:
5485                                         m_copyback(m, off,
5486                                             sizeof(struct icmp6_hdr),
5487                                             (caddr_t)pd->hdr.icmp6);
5488                                         m_copyback(m, ipoff2, sizeof(h2_6),
5489                                             (caddr_t)&h2_6);
5490                                         break;
5491 #endif /* INET6 */
5492                                 }
5493                                 m->m_flags &= ~M_HASH;
5494                                 m_copyback(m, off2, 8, (caddr_t)&th);
5495                         }
5496                         break;
5497                 }
5498                 case IPPROTO_UDP: {
5499                         struct udphdr           uh;
5500
5501                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
5502                             NULL, reason, pd2.af)) {
5503                                 DPFPRINTF(PF_DEBUG_MISC,
5504                                     ("pf: ICMP error message too short "
5505                                     "(udp)\n"));
5506                                 return (PF_DROP);
5507                         }
5508
5509                         key.af = pd2.af;
5510                         key.proto = IPPROTO_UDP;
5511                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5512                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5513                         key.port[pd2.sidx] = uh.uh_sport;
5514                         key.port[pd2.didx] = uh.uh_dport;
5515
5516                         STATE_LOOKUP(kif, &key, direction, *state, m);
5517                         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5518
5519                         /* translate source/destination address, if necessary */
5520                         if ((*state)->key[PF_SK_WIRE] !=
5521                             (*state)->key[PF_SK_STACK]) {
5522                                 struct pf_state_key *nk =
5523                                     (*state)->key[pd->didx];
5524
5525                                 if (PF_ANEQ(pd2.src,
5526                                     &nk->addr[pd2.sidx], pd2.af) ||
5527                                     nk->port[pd2.sidx] != uh.uh_sport)
5528                                         pf_change_icmp(pd2.src, &uh.uh_sport,
5529                                             daddr, &nk->addr[pd2.sidx],
5530                                             nk->port[pd2.sidx], &uh.uh_sum,
5531                                             pd2.ip_sum, icmpsum,
5532                                             pd->ip_sum, 1, pd2.af);
5533
5534                                 if (PF_ANEQ(pd2.dst,
5535                                     &nk->addr[pd2.didx], pd2.af) ||
5536                                     nk->port[pd2.didx] != uh.uh_dport)
5537                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
5538                                             NULL, /* XXX Inbound NAT? */
5539                                             &nk->addr[pd2.didx],
5540                                             nk->port[pd2.didx], &uh.uh_sum,
5541                                             pd2.ip_sum, icmpsum,
5542                                             pd->ip_sum, 1, pd2.af);
5543
5544                                 switch (pd2.af) {
5545 #ifdef INET
5546                                 case AF_INET:
5547                                         m_copyback(m, off, ICMP_MINLEN,
5548                                             (caddr_t)pd->hdr.icmp);
5549                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5550                                         break;
5551 #endif /* INET */
5552 #ifdef INET6
5553                                 case AF_INET6:
5554                                         m_copyback(m, off,
5555                                             sizeof(struct icmp6_hdr),
5556                                             (caddr_t)pd->hdr.icmp6);
5557                                         m_copyback(m, ipoff2, sizeof(h2_6),
5558                                             (caddr_t)&h2_6);
5559                                         break;
5560 #endif /* INET6 */
5561                                 }
5562                                 m->m_flags &= ~M_HASH;
5563                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
5564                         }
5565                         break;
5566                 }
5567 #ifdef INET
5568                 case IPPROTO_ICMP: {
5569                         struct icmp             iih;
5570
5571                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
5572                             NULL, reason, pd2.af)) {
5573                                 DPFPRINTF(PF_DEBUG_MISC,
5574                                     ("pf: ICMP error message too short i"
5575                                     "(icmp)\n"));
5576                                 return (PF_DROP);
5577                         }
5578
5579                         key.af = pd2.af;
5580                         key.proto = IPPROTO_ICMP;
5581                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5582                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5583                         key.port[0] = key.port[1] = iih.icmp_id;
5584
5585                         STATE_LOOKUP(kif, &key, direction, *state, m);
5586                         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5587
5588                         /* translate source/destination address, if necessary */
5589                         if ((*state)->key[PF_SK_WIRE] !=
5590                             (*state)->key[PF_SK_STACK]) {
5591                                 struct pf_state_key *nk =
5592                                     (*state)->key[pd->didx];
5593
5594                                 if (PF_ANEQ(pd2.src,
5595                                     &nk->addr[pd2.sidx], pd2.af) ||
5596                                     nk->port[pd2.sidx] != iih.icmp_id)
5597                                         pf_change_icmp(pd2.src, &iih.icmp_id,
5598                                             daddr, &nk->addr[pd2.sidx],
5599                                             nk->port[pd2.sidx], NULL,
5600                                             pd2.ip_sum, icmpsum,
5601                                             pd->ip_sum, 0, AF_INET);
5602
5603                                 if (PF_ANEQ(pd2.dst,
5604                                     &nk->addr[pd2.didx], pd2.af) ||
5605                                     nk->port[pd2.didx] != iih.icmp_id)
5606                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
5607                                             NULL, /* XXX Inbound NAT? */
5608                                             &nk->addr[pd2.didx],
5609                                             nk->port[pd2.didx], NULL,
5610                                             pd2.ip_sum, icmpsum,
5611                                             pd->ip_sum, 0, AF_INET);
5612
5613                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
5614                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5615                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5616                                 m->m_flags &= ~M_HASH;
5617                         }
5618                         break;
5619                 }
5620 #endif /* INET */
5621 #ifdef INET6
5622                 case IPPROTO_ICMPV6: {
5623                         struct icmp6_hdr        iih;
5624
5625                         if (!pf_pull_hdr(m, off2, &iih,
5626                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5627                                 DPFPRINTF(PF_DEBUG_MISC,
5628                                     ("pf: ICMP error message too short "
5629                                     "(icmp6)\n"));
5630                                 FAIL (PF_DROP);
5631                         }
5632
5633                         key.af = pd2.af;
5634                         key.proto = IPPROTO_ICMPV6;
5635                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5636                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5637                         key.port[0] = key.port[1] = iih.icmp6_id;
5638
5639                         STATE_LOOKUP(kif, &key, direction, *state, m);
5640                         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5641
5642                         /* translate source/destination address, if necessary */
5643                         if ((*state)->key[PF_SK_WIRE] !=
5644                             (*state)->key[PF_SK_STACK]) {
5645                                 struct pf_state_key *nk =
5646                                     (*state)->key[pd->didx];
5647
5648                                 if (PF_ANEQ(pd2.src,
5649                                     &nk->addr[pd2.sidx], pd2.af) ||
5650                                     nk->port[pd2.sidx] != iih.icmp6_id)
5651                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
5652                                             daddr, &nk->addr[pd2.sidx],
5653                                             nk->port[pd2.sidx], NULL,
5654                                             pd2.ip_sum, icmpsum,
5655                                             pd->ip_sum, 0, AF_INET6);
5656
5657                                 if (PF_ANEQ(pd2.dst,
5658                                     &nk->addr[pd2.didx], pd2.af) ||
5659                                     nk->port[pd2.didx] != iih.icmp6_id)
5660                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
5661                                             NULL, /* XXX Inbound NAT? */
5662                                             &nk->addr[pd2.didx],
5663                                             nk->port[pd2.didx], NULL,
5664                                             pd2.ip_sum, icmpsum,
5665                                             pd->ip_sum, 0, AF_INET6);
5666
5667                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
5668                                     (caddr_t)pd->hdr.icmp6);
5669                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5670                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5671                                     (caddr_t)&iih);
5672                                 m->m_flags &= ~M_HASH;
5673                         }
5674                         break;
5675                 }
5676 #endif /* INET6 */
5677                 default: {
5678                         key.af = pd2.af;
5679                         key.proto = pd2.proto;
5680                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5681                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5682                         key.port[0] = key.port[1] = 0;
5683
5684                         STATE_LOOKUP(kif, &key, direction, *state, m);
5685                         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5686
5687                         /* translate source/destination address, if necessary */
5688                         if ((*state)->key[PF_SK_WIRE] !=
5689                             (*state)->key[PF_SK_STACK]) {
5690                                 struct pf_state_key *nk =
5691                                     (*state)->key[pd->didx];
5692
5693                                 if (PF_ANEQ(pd2.src,
5694                                     &nk->addr[pd2.sidx], pd2.af))
5695                                         pf_change_icmp(pd2.src, NULL, daddr,
5696                                             &nk->addr[pd2.sidx], 0, NULL,
5697                                             pd2.ip_sum, icmpsum,
5698                                             pd->ip_sum, 0, pd2.af);
5699
5700                                 if (PF_ANEQ(pd2.dst,
5701                                     &nk->addr[pd2.didx], pd2.af))
5702                                         pf_change_icmp(pd2.src, NULL,
5703                                             NULL, /* XXX Inbound NAT? */
5704                                             &nk->addr[pd2.didx], 0, NULL,
5705                                             pd2.ip_sum, icmpsum,
5706                                             pd->ip_sum, 0, pd2.af);
5707
5708                                 switch (pd2.af) {
5709 #ifdef INET
5710                                 case AF_INET:
5711                                         m_copyback(m, off, ICMP_MINLEN,
5712                                             (caddr_t)pd->hdr.icmp);
5713                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5714                                         m->m_flags &= ~M_HASH;
5715                                         break;
5716 #endif /* INET */
5717 #ifdef INET6
5718                                 case AF_INET6:
5719                                         m_copyback(m, off,
5720                                             sizeof(struct icmp6_hdr),
5721                                             (caddr_t)pd->hdr.icmp6);
5722                                         m_copyback(m, ipoff2, sizeof(h2_6),
5723                                             (caddr_t)&h2_6);
5724                                         m->m_flags &= ~M_HASH;
5725                                         break;
5726 #endif /* INET6 */
5727                                 }
5728                         }
5729                         break;
5730                 }
5731                 }
5732         }
5733
5734         pfsync_update_state(*state);
5735         error = PF_PASS;
5736 done:
5737         if (*state)
5738                 lockmgr(&(*state)->lk, LK_RELEASE);
5739         return (error);
5740 }
5741
5742 /*
5743  * Test other connection state.  Caller must hold the state locked.
5744  */
5745 int
5746 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5747                     struct mbuf *m, struct pf_pdesc *pd)
5748 {
5749         struct pf_state_peer    *src, *dst;
5750         struct pf_state_key_cmp  key;
5751
5752         bzero(&key, sizeof(key));
5753         key.af = pd->af;
5754         key.proto = pd->proto;
5755         if (direction == PF_IN) {
5756                 PF_ACPY(&key.addr[0], pd->src, key.af);
5757                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5758                 key.port[0] = key.port[1] = 0;
5759         } else {
5760                 PF_ACPY(&key.addr[1], pd->src, key.af);
5761                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5762                 key.port[1] = key.port[0] = 0;
5763         }
5764
5765         STATE_LOOKUP(kif, &key, direction, *state, m);
5766         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5767
5768         if (direction == (*state)->direction) {
5769                 src = &(*state)->src;
5770                 dst = &(*state)->dst;
5771         } else {
5772                 src = &(*state)->dst;
5773                 dst = &(*state)->src;
5774         }
5775
5776         /* update states */
5777         if (src->state < PFOTHERS_SINGLE)
5778                 src->state = PFOTHERS_SINGLE;
5779         if (dst->state == PFOTHERS_SINGLE)
5780                 dst->state = PFOTHERS_MULTIPLE;
5781
5782         /* update expire time */
5783         (*state)->expire = time_second;
5784         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5785                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5786         else
5787                 (*state)->timeout = PFTM_OTHER_SINGLE;
5788
5789         /* translate source/destination address, if necessary */
5790         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5791                 struct pf_state_key *nk = (*state)->key[pd->didx];
5792
5793                 KKASSERT(nk);
5794                 KKASSERT(pd);
5795                 KKASSERT(pd->src);
5796                 KKASSERT(pd->dst);
5797                 switch (pd->af) {
5798 #ifdef INET
5799                 case AF_INET:
5800                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5801                                 pf_change_a(&pd->src->v4.s_addr,
5802                                     pd->ip_sum,
5803                                     nk->addr[pd->sidx].v4.s_addr,
5804                                     0);
5805
5806
5807                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5808                                 pf_change_a(&pd->dst->v4.s_addr,
5809                                     pd->ip_sum,
5810                                     nk->addr[pd->didx].v4.s_addr,
5811                                     0);
5812
5813                         break;
5814 #endif /* INET */
5815 #ifdef INET6
5816                 case AF_INET6:
5817                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5818                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5819
5820                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5821                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5822 #endif /* INET6 */
5823                 }
5824         }
5825
5826         pfsync_update_state(*state);
5827         lockmgr(&(*state)->lk, LK_RELEASE);
5828         return (PF_PASS);
5829 }
5830
5831 /*
5832  * ipoff and off are measured from the start of the mbuf chain.
5833  * h must be at "ipoff" on the mbuf chain.
5834  */
5835 void *
5836 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5837     u_short *actionp, u_short *reasonp, sa_family_t af)
5838 {
5839         switch (af) {
5840 #ifdef INET
5841         case AF_INET: {
5842                 struct ip       *h = mtod(m, struct ip *);
5843                 u_int16_t        fragoff = (h->ip_off & IP_OFFMASK) << 3;
5844
5845                 if (fragoff) {
5846                         if (fragoff >= len)
5847                                 ACTION_SET(actionp, PF_PASS);
5848                         else {
5849                                 ACTION_SET(actionp, PF_DROP);
5850                                 REASON_SET(reasonp, PFRES_FRAG);
5851                         }
5852                         return (NULL);
5853                 }
5854                 if (m->m_pkthdr.len < off + len ||
5855                     h->ip_len < off + len) {
5856                         ACTION_SET(actionp, PF_DROP);
5857                         REASON_SET(reasonp, PFRES_SHORT);
5858                         return (NULL);
5859                 }
5860                 break;
5861         }
5862 #endif /* INET */
5863 #ifdef INET6
5864         case AF_INET6: {
5865                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5866
5867                 if (m->m_pkthdr.len < off + len ||
5868                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5869                     (unsigned)(off + len)) {
5870                         ACTION_SET(actionp, PF_DROP);
5871                         REASON_SET(reasonp, PFRES_SHORT);
5872                         return (NULL);
5873                 }
5874                 break;
5875         }
5876 #endif /* INET6 */
5877         }
5878         m_copydata(m, off, len, p);
5879         return (p);
5880 }
5881
5882 int
5883 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif)
5884 {
5885         struct sockaddr_in      *dst;
5886         int                      ret = 1;
5887         int                      check_mpath;
5888 #ifdef INET6
5889         struct sockaddr_in6     *dst6;
5890         struct route_in6         ro;
5891 #else
5892         struct route             ro;
5893 #endif
5894         struct radix_node       *rn;
5895         struct rtentry          *rt;
5896         struct ifnet            *ifp;
5897
5898         check_mpath = 0;
5899         bzero(&ro, sizeof(ro));
5900         switch (af) {
5901         case AF_INET:
5902                 dst = satosin(&ro.ro_dst);
5903                 dst->sin_family = AF_INET;
5904                 dst->sin_len = sizeof(*dst);
5905                 dst->sin_addr = addr->v4;
5906                 break;
5907 #ifdef INET6
5908         case AF_INET6:
5909                 /*
5910                  * Skip check for addresses with embedded interface scope,
5911                  * as they would always match anyway.
5912                  */
5913                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5914                         goto out;
5915                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5916                 dst6->sin6_family = AF_INET6;
5917                 dst6->sin6_len = sizeof(*dst6);
5918                 dst6->sin6_addr = addr->v6;
5919                 break;
5920 #endif /* INET6 */
5921         default:
5922                 return (0);
5923         }
5924
5925         /* Skip checks for ipsec interfaces */
5926         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5927                 goto out;
5928
5929         rtalloc_ign((struct route *)&ro, 0);
5930
5931         if (ro.ro_rt != NULL) {
5932                 /* No interface given, this is a no-route check */
5933                 if (kif == NULL)
5934                         goto out;
5935
5936                 if (kif->pfik_ifp == NULL) {
5937                         ret = 0;
5938                         goto out;
5939                 }
5940
5941                 /* Perform uRPF check if passed input interface */
5942                 ret = 0;
5943                 rn = (struct radix_node *)ro.ro_rt;
5944                 do {
5945                         rt = (struct rtentry *)rn;
5946                         ifp = rt->rt_ifp;
5947
5948                         if (kif->pfik_ifp == ifp)
5949                                 ret = 1;
5950                         rn = NULL;
5951                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5952         } else
5953                 ret = 0;
5954 out:
5955         if (ro.ro_rt != NULL)
5956                 RTFREE(ro.ro_rt);
5957         return (ret);
5958 }
5959
5960 int
5961 pf_rtlabel_match(struct pf_addr *addr, sa_family_t af, struct pf_addr_wrap *aw)
5962 {
5963         struct sockaddr_in      *dst;
5964 #ifdef INET6
5965         struct sockaddr_in6     *dst6;
5966         struct route_in6         ro;
5967 #else
5968         struct route             ro;
5969 #endif
5970         int                      ret = 0;
5971
5972         ASSERT_LWKT_TOKEN_HELD(&pf_token);
5973
5974         bzero(&ro, sizeof(ro));
5975         switch (af) {
5976         case AF_INET:
5977                 dst = satosin(&ro.ro_dst);
5978                 dst->sin_family = AF_INET;
5979                 dst->sin_len = sizeof(*dst);
5980                 dst->sin_addr = addr->v4;
5981                 break;
5982 #ifdef INET6
5983         case AF_INET6:
5984                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5985                 dst6->sin6_family = AF_INET6;
5986                 dst6->sin6_len = sizeof(*dst6);
5987                 dst6->sin6_addr = addr->v6;
5988                 break;
5989 #endif /* INET6 */
5990         default:
5991                 return (0);
5992         }
5993
5994 rtalloc_ign((struct route *)&ro, (RTF_CLONING | RTF_PRCLONING));
5995
5996         if (ro.ro_rt != NULL) {
5997                 RTFREE(ro.ro_rt);
5998         }
5999
6000         return (ret);
6001 }
6002
6003 #ifdef INET
6004 void
6005 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
6006     struct pf_state *s, struct pf_pdesc *pd)
6007 {
6008         struct mbuf             *m0, *m1;
6009         struct route             iproute;
6010         struct route            *ro = NULL;
6011         struct sockaddr_in      *dst;
6012         struct ip               *ip;
6013         struct ifnet            *ifp = NULL;
6014         struct pf_addr           naddr;
6015         struct pf_src_node      *sn = NULL;
6016         int                      error = 0;
6017         int sw_csum;
6018
6019         ASSERT_LWKT_TOKEN_HELD(&pf_token);
6020
6021         if (m == NULL || *m == NULL || r == NULL ||
6022             (dir != PF_IN && dir != PF_OUT) || oifp == NULL)
6023                 panic("pf_route: invalid parameters");
6024
6025         if (((*m)->m_pkthdr.fw_flags & PF_MBUF_ROUTED) == 0) {
6026                 (*m)->m_pkthdr.fw_flags |= PF_MBUF_ROUTED;
6027                 (*m)->m_pkthdr.pf.routed = 1;
6028         } else {
6029                 if ((*m)->m_pkthdr.pf.routed++ > 3) {
6030                         m0 = *m;
6031                         *m = NULL;
6032                         goto bad;
6033                 }
6034         }
6035
6036         if (r->rt == PF_DUPTO) {
6037                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
6038                         return;
6039                 }
6040         } else {
6041                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
6042                         return;
6043                 }
6044                 m0 = *m;
6045         }
6046
6047         if (m0->m_len < sizeof(struct ip)) {
6048                 DPFPRINTF(PF_DEBUG_URGENT,
6049                     ("pf_route: m0->m_len < sizeof(struct ip)\n"));
6050                 goto bad;
6051         }
6052
6053         ip = mtod(m0, struct ip *);
6054
6055         ro = &iproute;
6056         bzero((caddr_t)ro, sizeof(*ro));
6057         dst = satosin(&ro->ro_dst);
6058         dst->sin_family = AF_INET;
6059         dst->sin_len = sizeof(*dst);
6060         dst->sin_addr = ip->ip_dst;
6061
6062         if (r->rt == PF_FASTROUTE) {
6063                 rtalloc(ro);
6064                 if (ro->ro_rt == 0) {
6065                         ipstat.ips_noroute++;
6066                         goto bad;
6067                 }
6068
6069                 ifp = ro->ro_rt->rt_ifp;
6070                 ro->ro_rt->rt_use++;
6071
6072                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
6073                         dst = satosin(ro->ro_rt->rt_gateway);
6074         } else {
6075                 if (TAILQ_EMPTY(&r->rpool.list)) {
6076                         DPFPRINTF(PF_DEBUG_URGENT,
6077                             ("pf_route: TAILQ_EMPTY(&r->rpool.list)\n"));
6078                         goto bad;
6079                 }
6080                 if (s == NULL) {
6081                         pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
6082                             &naddr, NULL, &sn);
6083                         if (!PF_AZERO(&naddr, AF_INET))
6084                                 dst->sin_addr.s_addr = naddr.v4.s_addr;
6085                         ifp = r->rpool.cur->kif ?
6086                             r->rpool.cur->kif->pfik_ifp : NULL;
6087                 } else {
6088                         if (!PF_AZERO(&s->rt_addr, AF_INET))
6089                                 dst->sin_addr.s_addr =
6090                                     s->rt_addr.v4.s_addr;
6091                         ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
6092                 }
6093         }
6094         if (ifp == NULL)
6095                 goto bad;
6096
6097         if (oifp != ifp) {
6098                 if (pf_test(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) {
6099                         goto bad;
6100                 } else if (m0 == NULL) {
6101                         goto done;
6102                 }
6103                 if (m0->m_len < sizeof(struct ip)) {
6104                         DPFPRINTF(PF_DEBUG_URGENT,
6105                             ("pf_route: m0->m_len < sizeof(struct ip)\n"));
6106                         goto bad;
6107                 }
6108                 ip = mtod(m0, struct ip *);
6109         }
6110
6111         /* Copied from FreeBSD 5.1-CURRENT ip_output. */
6112         m0->m_pkthdr.csum_flags |= CSUM_IP;
6113         sw_csum = m0->m_pkthdr.csum_flags & ~ifp->if_hwassist;
6114         if (sw_csum & CSUM_DELAY_DATA) {
6115                 in_delayed_cksum(m0);
6116                 sw_csum &= ~CSUM_DELAY_DATA;
6117         }
6118         m0->m_pkthdr.csum_flags &= ifp->if_hwassist;
6119         m0->m_pkthdr.csum_iphlen = (ip->ip_hl << 2);
6120
6121         /*
6122          * WARNING!  We cannot fragment if the packet was modified from an
6123          *           original which expected to be using TSO.  In this
6124          *           situation we pray that the target interface is
6125          *           compatible with the originating interface.
6126          */
6127         if (ip->ip_len <= ifp->if_mtu ||
6128             (m0->m_pkthdr.csum_flags & CSUM_TSO) ||
6129             ((ifp->if_hwassist & CSUM_FRAGMENT) &&
6130                 (ip->ip_off & IP_DF) == 0)) {
6131                 ip->ip_len = htons(ip->ip_len);
6132                 ip->ip_off = htons(ip->ip_off);
6133                 ip->ip_sum = 0;
6134                 if (sw_csum & CSUM_DELAY_IP) {
6135                         /* From KAME */
6136                         if (ip->ip_v == IPVERSION &&
6137                             (ip->ip_hl << 2) == sizeof(*ip)) {
6138                                 ip->ip_sum = in_cksum_hdr(ip);
6139                         } else {
6140                                 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
6141                         }
6142                 }
6143                 lwkt_reltoken(&pf_token);
6144                 error = ifp->if_output(ifp, m0, sintosa(dst), ro->ro_rt);
6145                 lwkt_gettoken(&pf_token);
6146                 goto done;
6147         }
6148
6149         /*
6150          * Too large for interface; fragment if possible.
6151          * Must be able to put at least 8 bytes per fragment.
6152          */
6153         if (ip->ip_off & IP_DF) {
6154                 ipstat.ips_cantfrag++;
6155                 if (r->rt != PF_DUPTO) {
6156                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
6157                                    ifp->if_mtu);
6158                         goto done;
6159                 } else
6160                         goto bad;
6161         }
6162
6163         m1 = m0;
6164         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist, sw_csum);
6165         if (error) {
6166                 goto bad;
6167         }
6168
6169         for (m0 = m1; m0; m0 = m1) {
6170                 m1 = m0->m_nextpkt;
6171                 m0->m_nextpkt = 0;
6172                 if (error == 0) {
6173                         lwkt_reltoken(&pf_token);
6174                         error = (*ifp->if_output)(ifp, m0, sintosa(dst),
6175                                                   NULL);
6176                         lwkt_gettoken(&pf_token);
6177                 } else
6178                         m_freem(m0);
6179         }
6180
6181         if (error == 0)
6182                 ipstat.ips_fragmented++;
6183
6184 done:
6185         if (r->rt != PF_DUPTO)
6186                 *m = NULL;
6187         if (ro == &iproute && ro->ro_rt)
6188                 RTFREE(ro->ro_rt);
6189         return;
6190
6191 bad:
6192         m_freem(m0);
6193         goto done;
6194 }
6195 #endif /* INET */
6196
6197 #ifdef INET6
6198 void
6199 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
6200     struct pf_state *s, struct pf_pdesc *pd)
6201 {
6202         struct mbuf             *m0;
6203         struct route_in6         ip6route;
6204         struct route_in6        *ro;
6205         struct sockaddr_in6     *dst;
6206         struct ip6_hdr          *ip6;
6207         struct ifnet            *ifp = NULL;
6208         struct pf_addr           naddr;
6209         struct pf_src_node      *sn = NULL;
6210
6211         if (m == NULL || *m == NULL || r == NULL ||
6212             (dir != PF_IN && dir != PF_OUT) || oifp == NULL)
6213                 panic("pf_route6: invalid parameters");
6214
6215         if (((*m)->m_pkthdr.fw_flags & PF_MBUF_ROUTED) == 0) {
6216                 (*m)->m_pkthdr.fw_flags |= PF_MBUF_ROUTED;
6217                 (*m)->m_pkthdr.pf.routed = 1;
6218         } else {
6219                 if ((*m)->m_pkthdr.pf.routed++ > 3) {
6220                         m0 = *m;
6221                         *m = NULL;
6222                         goto bad;
6223                 }
6224         }
6225
6226         if (r->rt == PF_DUPTO) {
6227                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL)
6228                         return;
6229         } else {
6230                 if ((r->rt == PF_REPLYTO) == (r->direction == dir))
6231                         return;
6232                 m0 = *m;
6233         }
6234
6235         if (m0->m_len < sizeof(struct ip6_hdr)) {
6236                 DPFPRINTF(PF_DEBUG_URGENT,
6237                     ("pf_route6: m0->m_len < sizeof(struct ip6_hdr)\n"));
6238                 goto bad;
6239         }
6240         ip6 = mtod(m0, struct ip6_hdr *);
6241
6242         ro = &ip6route;
6243         bzero((caddr_t)ro, sizeof(*ro));
6244         dst = (struct sockaddr_in6 *)&ro->ro_dst;
6245         dst->sin6_family = AF_INET6;
6246         dst->sin6_len = sizeof(*dst);
6247         dst->sin6_addr = ip6->ip6_dst;
6248
6249         /*
6250          * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6251          * so make sure pf.flags is clear.
6252          *
6253          * Cheat. XXX why only in the v6 case???
6254          */
6255         if (r->rt == PF_FASTROUTE) {
6256                 m0->m_pkthdr.fw_flags |= PF_MBUF_TAGGED;
6257                 m0->m_pkthdr.pf.flags = 0;
6258                 /* XXX Re-Check when Upgrading to > 4.4 */
6259                 m0->m_pkthdr.pf.statekey = NULL;
6260                 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
6261                 return;
6262         }
6263
6264         if (TAILQ_EMPTY(&r->rpool.list)) {
6265                 DPFPRINTF(PF_DEBUG_URGENT,
6266                     ("pf_route6: TAILQ_EMPTY(&r->rpool.list)\n"));
6267                 goto bad;
6268         }
6269         if (s == NULL) {
6270                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
6271                     &naddr, NULL, &sn);
6272                 if (!PF_AZERO(&naddr, AF_INET6))
6273                         PF_ACPY((struct pf_addr *)&dst->sin6_addr,
6274                             &naddr, AF_INET6);
6275                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
6276         } else {
6277                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
6278                         PF_ACPY((struct pf_addr *)&dst->sin6_addr,
6279                             &s->rt_addr, AF_INET6);
6280                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
6281         }
6282         if (ifp == NULL)
6283                 goto bad;
6284
6285         if (oifp != ifp) {
6286                 if (pf_test6(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) {
6287                         goto bad;
6288                 } else if (m0 == NULL) {
6289                         goto done;
6290                 }
6291                 if (m0->m_len < sizeof(struct ip6_hdr)) {
6292                         DPFPRINTF(PF_DEBUG_URGENT,
6293                             ("pf_route6: m0->m_len < sizeof(struct ip6_hdr)\n"));
6294                         goto bad;
6295                 }
6296                 ip6 = mtod(m0, struct ip6_hdr *);
6297         }
6298
6299         /*
6300          * If the packet is too large for the outgoing interface,
6301          * send back an icmp6 error.
6302          */
6303         if (IN6_IS_SCOPE_EMBED(&dst->sin6_addr))
6304                 dst->sin6_addr.s6_addr16[1] = htons(ifp->if_index);
6305         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) {
6306                 nd6_output(ifp, ifp, m0, dst, NULL);
6307         } else {
6308                 in6_ifstat_inc(ifp, ifs6_in_toobig);
6309                 if (r->rt != PF_DUPTO)
6310                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
6311                 else
6312                         goto bad;
6313         }
6314
6315 done:
6316         if (r->rt != PF_DUPTO)
6317                 *m = NULL;
6318         return;
6319
6320 bad:
6321         m_freem(m0);
6322         goto done;
6323 }
6324 #endif /* INET6 */
6325
6326
6327 /*
6328  * check protocol (tcp/udp/icmp/icmp6) checksum and set mbuf flag
6329  *   off is the offset where the protocol header starts
6330  *   len is the total length of protocol header plus payload
6331  * returns 0 when the checksum is valid, otherwise returns 1.
6332  */
6333 /*
6334  * XXX
6335  * FreeBSD supports cksum offload for the following drivers.
6336  * em(4), gx(4), lge(4), nge(4), ti(4), xl(4)
6337  * If we can make full use of it we would outperform ipfw/ipfilter in
6338  * very heavy traffic.
6339  * I have not tested 'cause I don't have NICs that supports cksum offload.
6340  * (There might be problems. Typical phenomena would be
6341  *   1. No route message for UDP packet.
6342  *   2. No connection acceptance from external hosts regardless of rule set.)
6343  */
6344 int
6345 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p,
6346     sa_family_t af)
6347 {
6348         u_int16_t sum = 0;
6349         int hw_assist = 0;
6350         struct ip *ip;
6351
6352         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
6353                 return (1);
6354         if (m->m_pkthdr.len < off + len)
6355                 return (1);
6356
6357         switch (p) {
6358         case IPPROTO_TCP:
6359         case IPPROTO_UDP:
6360                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
6361                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
6362                                 sum = m->m_pkthdr.csum_data;
6363                         } else {
6364                                 ip = mtod(m, struct ip *);
6365                                 sum = in_pseudo(ip->ip_src.s_addr,
6366                                         ip->ip_dst.s_addr, htonl((u_short)len +
6367                                         m->m_pkthdr.csum_data + p));
6368                         }
6369                         sum ^= 0xffff;
6370                         ++hw_assist;
6371                 }
6372                 break;
6373         case IPPROTO_ICMP:
6374 #ifdef INET6
6375         case IPPROTO_ICMPV6:
6376 #endif /* INET6 */
6377                 break;
6378         default:
6379                 return (1);
6380         }
6381
6382         if (!hw_assist) {
6383                 switch (af) {
6384                 case AF_INET:
6385                         if (p == IPPROTO_ICMP) {
6386                                 if (m->m_len < off)
6387                                         return (1);
6388                                 m->m_data += off;
6389                                 m->m_len -= off;
6390                                 sum = in_cksum(m, len);
6391                                 m->m_data -= off;
6392                                 m->m_len += off;
6393                         } else {
6394                                 if (m->m_len < sizeof(struct ip))
6395                                         return (1);
6396                                 sum = in_cksum_range(m, p, off, len);
6397                                 if (sum == 0) {
6398                                         m->m_pkthdr.csum_flags |=
6399                                             (CSUM_DATA_VALID |
6400                                              CSUM_PSEUDO_HDR);
6401                                         m->m_pkthdr.csum_data = 0xffff;
6402                                 }
6403                         }
6404                         break;
6405 #ifdef INET6
6406                 case AF_INET6:
6407                         if (m->m_len < sizeof(struct ip6_hdr))
6408                                 return (1);
6409                         sum = in6_cksum(m, p, off, len);
6410                         /*
6411                          * XXX
6412                          * IPv6 H/W cksum off-load not supported yet!
6413                          *
6414                          * if (sum == 0) {
6415                          *      m->m_pkthdr.csum_flags |=
6416                          *          (CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
6417                          *      m->m_pkthdr.csum_data = 0xffff;
6418                          *}
6419                          */
6420                         break;
6421 #endif /* INET6 */
6422                 default:
6423                         return (1);
6424                 }
6425         }
6426         if (sum) {
6427                 switch (p) {
6428                 case IPPROTO_TCP:
6429                         tcpstat.tcps_rcvbadsum++;
6430                         break;
6431                 case IPPROTO_UDP:
6432                         udp_stat.udps_badsum++;
6433                         break;
6434                 case IPPROTO_ICMP:
6435                         icmpstat.icps_checksum++;
6436                         break;
6437 #ifdef INET6
6438                 case IPPROTO_ICMPV6:
6439                         icmp6stat.icp6s_checksum++;
6440                         break;
6441 #endif /* INET6 */
6442                 }
6443                 return (1);
6444         }
6445         return (0);
6446 }
6447
6448 struct pf_divert *
6449 pf_find_divert(struct mbuf *m)
6450 {
6451         struct m_tag    *mtag;
6452
6453         if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL)
6454                 return (NULL);
6455
6456         return ((struct pf_divert *)(mtag + 1));
6457 }
6458
6459 struct pf_divert *
6460 pf_get_divert(struct mbuf *m)
6461 {
6462         struct m_tag    *mtag;
6463
6464         if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) {
6465                 mtag = m_tag_get(PACKET_TAG_PF_DIVERT, sizeof(struct pf_divert),
6466                     M_NOWAIT);
6467                 if (mtag == NULL)
6468                         return (NULL);
6469                 bzero(mtag + 1, sizeof(struct pf_divert));
6470                 m_tag_prepend(m, mtag);
6471         }
6472
6473         return ((struct pf_divert *)(mtag + 1));
6474 }
6475
6476 #ifdef INET
6477
6478 /*
6479  * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE
6480  */
6481 int
6482 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0,
6483     struct ether_header *eh, struct inpcb *inp)
6484 {
6485         struct pfi_kif          *kif;
6486         u_short                  action, reason = 0, log = 0;
6487         struct mbuf             *m = *m0;
6488         struct ip               *h = NULL;
6489         struct pf_rule          *a = NULL, *r = &pf_default_rule, *tr, *nr;
6490         struct pf_state         *s = NULL;
6491         struct pf_ruleset       *ruleset = NULL;
6492         struct pf_pdesc          pd;
6493         int                      off, dirndx;
6494 #ifdef ALTQ
6495         int                      pqid = 0;
6496 #endif
6497
6498         if (m->m_pkthdr.fw_flags & IPFW_MBUF_CONTINUE) {
6499                 /* Skip us; continue in ipfw. */
6500                 return (PF_PASS);
6501         }
6502
6503         if (!pf_status.running)
6504                 return (PF_PASS);
6505
6506         memset(&pd, 0, sizeof(pd));
6507 #ifdef foo
6508         if (ifp->if_type == IFT_CARP && ifp->if_carpdev)
6509                 kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif;
6510         else
6511 #endif
6512                 kif = (struct pfi_kif *)ifp->if_pf_kif;
6513
6514         if (kif == NULL) {
6515                 DPFPRINTF(PF_DEBUG_URGENT,
6516                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
6517                 return (PF_DROP);
6518         }
6519         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6520                 return (PF_PASS);
6521
6522 #ifdef DIAGNOSTIC
6523         if ((m->m_flags & M_PKTHDR) == 0)
6524                 panic("non-M_PKTHDR is passed to pf_test");
6525 #endif /* DIAGNOSTIC */
6526
6527         if (m->m_pkthdr.len < (int)sizeof(*h)) {
6528                 action = PF_DROP;
6529                 REASON_SET(&reason, PFRES_SHORT);
6530                 log = 1;
6531                 goto done;
6532         }
6533
6534         /*
6535          * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6536          * so make sure pf.flags is clear.
6537          */
6538         if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED)
6539                 return (PF_PASS);
6540         m->m_pkthdr.pf.flags = 0;
6541         /* Re-Check when updating to > 4.4 */
6542         m->m_pkthdr.pf.statekey = NULL;
6543
6544         /* We do IP header normalization and packet reassembly here */
6545         if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
6546                 action = PF_DROP;
6547                 goto done;
6548         }
6549         m = *m0;        /* pf_normalize messes with m0 */
6550         h = mtod(m, struct ip *);
6551
6552         off = h->ip_hl << 2;
6553         if (off < (int)sizeof(*h)) {
6554                 action = PF_DROP;
6555                 REASON_SET(&reason, PFRES_SHORT);
6556                 log = 1;
6557                 goto done;
6558         }
6559
6560         pd.src = (struct pf_addr *)&h->ip_src;
6561         pd.dst = (struct pf_addr *)&h->ip_dst;
6562         pd.sport = pd.dport = NULL;
6563         pd.ip_sum = &h->ip_sum;
6564         pd.proto_sum = NULL;
6565         pd.proto = h->ip_p;
6566         pd.dir = dir;
6567         pd.sidx = (dir == PF_IN) ? 0 : 1;
6568         pd.didx = (dir == PF_IN) ? 1 : 0;
6569         pd.af = AF_INET;
6570         pd.tos = h->ip_tos;
6571         pd.tot_len = h->ip_len;
6572         pd.eh = eh;
6573
6574         /* handle fragments that didn't get reassembled by normalization */
6575         if (h->ip_off & (IP_MF | IP_OFFMASK)) {
6576                 action = pf_test_fragment(&r, dir, kif, m, h,
6577                     &pd, &a, &ruleset);
6578                 goto done;
6579         }
6580
6581         switch (h->ip_p) {
6582
6583         case IPPROTO_TCP: {
6584                 struct tcphdr   th;
6585
6586                 pd.hdr.tcp = &th;
6587                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6588                     &action, &reason, AF_INET)) {
6589                         log = action != PF_PASS;
6590                         goto done;
6591                 }
6592                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6593 #ifdef ALTQ
6594                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
6595                         pqid = 1;
6596 #endif
6597                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6598                 if (action == PF_DROP)
6599                         goto done;
6600                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6601                                            &reason);
6602                 if (action == PF_PASS) {
6603                         r = s->rule.ptr;
6604                         a = s->anchor.ptr;
6605                         log = s->log;
6606                 } else if (s == NULL) {
6607                         action = pf_test_rule(&r, &s, dir, kif,
6608                                               m, off, h, &pd, &a,
6609                                               &ruleset, NULL, inp);
6610                 }
6611                 break;
6612         }
6613
6614         case IPPROTO_UDP: {
6615                 struct udphdr   uh;
6616
6617                 pd.hdr.udp = &uh;
6618                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6619                     &action, &reason, AF_INET)) {
6620                         log = action != PF_PASS;
6621                         goto done;
6622                 }
6623                 if (uh.uh_dport == 0 ||
6624                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6625                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6626                         action = PF_DROP;
6627                         REASON_SET(&reason, PFRES_SHORT);
6628                         goto done;
6629                 }
6630                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6631                 if (action == PF_PASS) {
6632                         r = s->rule.ptr;
6633                         a = s->anchor.ptr;
6634                         log = s->log;
6635                 } else if (s == NULL) {
6636                         action = pf_test_rule(&r, &s, dir, kif,
6637                                               m, off, h, &pd, &a,
6638                                               &ruleset, NULL, inp);
6639                 }
6640                 break;
6641         }
6642
6643         case IPPROTO_ICMP: {
6644                 struct icmp     ih;
6645
6646                 pd.hdr.icmp = &ih;
6647                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
6648                     &action, &reason, AF_INET)) {
6649                         log = action != PF_PASS;
6650                         goto done;
6651                 }
6652                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6653                                             &reason);
6654                 if (action == PF_PASS) {
6655                         r = s->rule.ptr;
6656                         a = s->anchor.ptr;
6657                         log = s->log;
6658                 } else if (s == NULL) {
6659                         action = pf_test_rule(&r, &s, dir, kif,
6660                                               m, off, h, &pd, &a,
6661                                               &ruleset, NULL, inp);
6662                 }
6663                 break;
6664         }
6665
6666         default:
6667                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6668                 if (action == PF_PASS) {
6669                         r = s->rule.ptr;
6670                         a = s->anchor.ptr;
6671                         log = s->log;
6672                 } else if (s == NULL) {
6673                         action = pf_test_rule(&r, &s, dir, kif, m, off, h,
6674                                               &pd, &a, &ruleset, NULL, inp);
6675                 }
6676                 break;
6677         }
6678
6679 done:
6680         if (action == PF_PASS && h->ip_hl > 5 &&
6681             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6682                 action = PF_DROP;
6683                 REASON_SET(&reason, PFRES_IPOPTIONS);
6684                 log = 1;
6685                 DPFPRINTF(PF_DEBUG_MISC,
6686                     ("pf: dropping packet with ip options\n"));
6687         }
6688
6689         if ((s && s->tag) || r->rtableid)
6690                 pf_tag_packet(m, s ? s->tag : 0, r->rtableid);
6691
6692 #if 0
6693         if (dir == PF_IN && s && s->key[PF_SK_STACK])
6694                 m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK];
6695 #endif
6696
6697 #ifdef ALTQ
6698         /*
6699          * Generate a hash code and qid request for ALTQ.  A qid of 0
6700          * is allowed and will cause altq to select the default queue.
6701          */
6702         if (action == PF_PASS) {
6703                 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
6704                 if (pqid || (pd.tos & IPTOS_LOWDELAY))
6705                         m->m_pkthdr.pf.qid = r->pqid;
6706                 else
6707                         m->m_pkthdr.pf.qid = r->qid;
6708                 m->m_pkthdr.pf.ecn_af = AF_INET;
6709                 m->m_pkthdr.pf.hdr = h;
6710                 /* add connection hash for fairq */
6711                 if (s) {
6712                         /* for fairq */
6713                         m->m_pkthdr.pf.state_hash = s->hash;
6714                         m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED;
6715                 }
6716         }
6717 #endif /* ALTQ */
6718
6719         /*
6720          * connections redirected to loopback should not match sockets
6721          * bound specifically to loopback due to security implications,
6722          * see tcp_input() and in_pcblookup_listen().
6723          */
6724         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6725             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6726             (s->nat_rule.ptr->action == PF_RDR ||
6727             s->nat_rule.ptr->action == PF_BINAT) &&
6728             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
6729                 m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST;
6730
6731         if (dir == PF_IN && action == PF_PASS && r->divert.port) {
6732                 struct pf_divert *divert;
6733
6734                 if ((divert = pf_get_divert(m))) {
6735                         m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED;
6736                         divert->port = r->divert.port;
6737                         divert->addr.ipv4 = r->divert.addr.v4;
6738                 }
6739         }
6740
6741         if (log) {
6742                 struct pf_rule *lr;
6743
6744                 if (s != NULL && s->nat_rule.ptr != NULL &&
6745                     s->nat_rule.ptr->log & PF_LOG_ALL)
6746                         lr = s->nat_rule.ptr;
6747                 else
6748                         lr = r;
6749                 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, lr, a, ruleset,
6750                     &pd);
6751         }
6752
6753         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6754         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6755
6756         if (action == PF_PASS || r->action == PF_DROP) {
6757                 dirndx = (dir == PF_OUT);
6758                 r->packets[dirndx]++;
6759                 r->bytes[dirndx] += pd.tot_len;
6760                 if (a != NULL) {
6761                         a->packets[dirndx]++;
6762                         a->bytes[dirndx] += pd.tot_len;
6763                 }
6764                 if (s != NULL) {
6765                         if (s->nat_rule.ptr != NULL) {
6766                                 s->nat_rule.ptr->packets[dirndx]++;
6767                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6768                         }
6769                         if (s->src_node != NULL) {
6770                                 s->src_node->packets[dirndx]++;
6771                                 s->src_node->bytes[dirndx] += pd.tot_len;
6772                         }
6773                         if (s->nat_src_node != NULL) {
6774                                 s->nat_src_node->packets[dirndx]++;
6775                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6776                         }
6777                         dirndx = (dir == s->direction) ? 0 : 1;
6778                         s->packets[dirndx]++;
6779                         s->bytes[dirndx] += pd.tot_len;
6780                 }
6781                 tr = r;
6782                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6783                 if (nr != NULL && r == &pf_default_rule)
6784                         tr = nr;
6785                 if (tr->src.addr.type == PF_ADDR_TABLE)
6786                         pfr_update_stats(tr->src.addr.p.tbl,
6787                             (s == NULL) ? pd.src :
6788                             &s->key[(s->direction == PF_IN)]->
6789                                 addr[(s->direction == PF_OUT)],
6790                             pd.af, pd.tot_len, dir == PF_OUT,
6791                             r->action == PF_PASS, tr->src.neg);
6792                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6793                         pfr_update_stats(tr->dst.addr.p.tbl,
6794                             (s == NULL) ? pd.dst :
6795                             &s->key[(s->direction == PF_IN)]->
6796                                 addr[(s->direction == PF_IN)],
6797                             pd.af, pd.tot_len, dir == PF_OUT,
6798                             r->action == PF_PASS, tr->dst.neg);
6799         }
6800
6801
6802         if (action == PF_SYNPROXY_DROP) {
6803                 m_freem(*m0);
6804                 *m0 = NULL;
6805                 action = PF_PASS;
6806         } else if (r->rt) {
6807                 /* pf_route can free the mbuf causing *m0 to become NULL */
6808                 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6809         }
6810
6811         return (action);
6812 }
6813 #endif /* INET */
6814
6815 #ifdef INET6
6816
6817 /*
6818  * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE
6819  */
6820 int
6821 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0,
6822     struct ether_header *eh, struct inpcb *inp)
6823 {
6824         struct pfi_kif          *kif;
6825         u_short                  action, reason = 0, log = 0;
6826         struct mbuf             *m = *m0, *n = NULL;
6827         struct ip6_hdr          *h = NULL;
6828         struct pf_rule          *a = NULL, *r = &pf_default_rule, *tr, *nr;
6829         struct pf_state         *s = NULL;
6830         struct pf_ruleset       *ruleset = NULL;
6831         struct pf_pdesc          pd;
6832         int                      off, terminal = 0, dirndx, rh_cnt = 0;
6833
6834         if (!pf_status.running)
6835                 return (PF_PASS);
6836
6837         memset(&pd, 0, sizeof(pd));
6838 #ifdef foo
6839         if (ifp->if_type == IFT_CARP && ifp->if_carpdev)
6840                 kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif;
6841         else
6842 #endif
6843                 kif = (struct pfi_kif *)ifp->if_pf_kif;
6844
6845         if (kif == NULL) {
6846                 DPFPRINTF(PF_DEBUG_URGENT,
6847                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6848                 return (PF_DROP);
6849         }
6850         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6851                 return (PF_PASS);
6852
6853 #ifdef DIAGNOSTIC
6854         if ((m->m_flags & M_PKTHDR) == 0)
6855                 panic("non-M_PKTHDR is passed to pf_test6");
6856 #endif /* DIAGNOSTIC */
6857
6858         if (m->m_pkthdr.len < (int)sizeof(*h)) {
6859                 action = PF_DROP;
6860                 REASON_SET(&reason, PFRES_SHORT);
6861                 log = 1;
6862                 goto done;
6863         }
6864
6865         /*
6866          * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6867          * so make sure pf.flags is clear.
6868          */
6869         if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED)
6870                 return (PF_PASS);
6871         m->m_pkthdr.pf.flags = 0;
6872         /* Re-Check when updating to > 4.4 */
6873         m->m_pkthdr.pf.statekey = NULL;
6874
6875         /* We do IP header normalization and packet reassembly here */
6876         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6877                 action = PF_DROP;
6878                 goto done;
6879         }
6880         m = *m0;        /* pf_normalize messes with m0 */
6881         h = mtod(m, struct ip6_hdr *);
6882
6883 #if 1
6884         /*
6885          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6886          * will do something bad, so drop the packet for now.
6887          */
6888         if (htons(h->ip6_plen) == 0) {
6889                 action = PF_DROP;
6890                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6891                 goto done;
6892         }
6893 #endif
6894
6895         pd.src = (struct pf_addr *)&h->ip6_src;
6896         pd.dst = (struct pf_addr *)&h->ip6_dst;
6897         pd.sport = pd.dport = NULL;
6898         pd.ip_sum = NULL;
6899         pd.proto_sum = NULL;
6900         pd.dir = dir;
6901         pd.sidx = (dir == PF_IN) ? 0 : 1;
6902         pd.didx = (dir == PF_IN) ? 1 : 0;
6903         pd.af = AF_INET6;
6904         pd.tos = 0;
6905         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6906         pd.eh = eh;
6907
6908         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6909         pd.proto = h->ip6_nxt;
6910         do {
6911                 switch (pd.proto) {
6912                 case IPPROTO_FRAGMENT:
6913                         action = pf_test_fragment(&r, dir, kif, m, h,
6914                             &pd, &a, &ruleset);
6915                         if (action == PF_DROP)
6916                                 REASON_SET(&reason, PFRES_FRAG);
6917                         goto done;
6918                 case IPPROTO_ROUTING: {
6919                         struct ip6_rthdr rthdr;
6920
6921                         if (rh_cnt++) {
6922                                 DPFPRINTF(PF_DEBUG_MISC,
6923                                     ("pf: IPv6 more than one rthdr\n"));
6924                                 action = PF_DROP;
6925                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6926                                 log = 1;
6927                                 goto done;
6928                         }
6929                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6930                             &reason, pd.af)) {
6931                                 DPFPRINTF(PF_DEBUG_MISC,
6932                                     ("pf: IPv6 short rthdr\n"));
6933                                 action = PF_DROP;
6934                                 REASON_SET(&reason, PFRES_SHORT);
6935                                 log = 1;
6936                                 goto done;
6937                         }
6938                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6939                                 DPFPRINTF(PF_DEBUG_MISC,
6940                                     ("pf: IPv6 rthdr0\n"));
6941                                 action = PF_DROP;
6942                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6943                                 log = 1;
6944                                 goto done;
6945                         }
6946                         /* FALLTHROUGH */
6947                 }
6948                 case IPPROTO_AH:
6949                 case IPPROTO_HOPOPTS:
6950                 case IPPROTO_DSTOPTS: {
6951                         /* get next header and header length */
6952                         struct ip6_ext  opt6;
6953
6954                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6955                             NULL, &reason, pd.af)) {
6956                                 DPFPRINTF(PF_DEBUG_MISC,
6957                                     ("pf: IPv6 short opt\n"));
6958                                 action = PF_DROP;
6959                                 log = 1;
6960                                 goto done;
6961                         }
6962                         if (pd.proto == IPPROTO_AH)
6963                                 off += (opt6.ip6e_len + 2) * 4;
6964                         else
6965                                 off += (opt6.ip6e_len + 1) * 8;
6966                         pd.proto = opt6.ip6e_nxt;
6967                         /* goto the next header */
6968                         break;
6969                 }
6970                 default:
6971                         terminal++;
6972                         break;
6973                 }
6974         } while (!terminal);
6975
6976         /* if there's no routing header, use unmodified mbuf for checksumming */
6977         if (!n)
6978                 n = m;
6979
6980         switch (pd.proto) {
6981
6982         case IPPROTO_TCP: {
6983                 struct tcphdr   th;
6984
6985                 pd.hdr.tcp = &th;
6986                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6987                     &action, &reason, AF_INET6)) {
6988                         log = action != PF_PASS;
6989                         goto done;
6990                 }
6991                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6992                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6993                 if (action == PF_DROP)
6994                         goto done;
6995                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6996                                            &reason);
6997                 if (action == PF_PASS) {
6998                         r = s->rule.ptr;
6999                         a = s->anchor.ptr;
7000                         log = s->log;
7001                 } else if (s == NULL) {
7002                         action = pf_test_rule(&r, &s, dir, kif,
7003                                               m, off, h, &pd, &a,
7004                                               &ruleset, NULL, inp);
7005                 }
7006                 break;
7007         }
7008
7009         case IPPROTO_UDP: {
7010                 struct udphdr   uh;
7011
7012                 pd.hdr.udp = &uh;
7013                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
7014                     &action, &reason, AF_INET6)) {
7015                         log = action != PF_PASS;
7016                         goto done;
7017                 }
7018                 if (uh.uh_dport == 0 ||
7019                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
7020                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
7021                         action = PF_DROP;
7022                         REASON_SET(&reason, PFRES_SHORT);
7023                         goto done;
7024                 }
7025                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
7026                 if (action == PF_PASS) {
7027                         r = s->rule.ptr;
7028                         a = s->anchor.ptr;
7029                         log = s->log;
7030                 } else if (s == NULL) {
7031                         action = pf_test_rule(&r, &s, dir, kif,
7032                                               m, off, h, &pd, &a,
7033                                               &ruleset, NULL, inp);
7034                 }
7035                 break;
7036         }
7037
7038         case IPPROTO_ICMPV6: {
7039                 struct icmp6_hdr        ih;
7040
7041                 pd.hdr.icmp6 = &ih;
7042                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
7043                     &action, &reason, AF_INET6)) {
7044                         log = action != PF_PASS;
7045                         goto done;
7046                 }
7047                 action = pf_test_state_icmp(&s, dir, kif,
7048                                             m, off, h, &pd, &reason);
7049                 if (action == PF_PASS) {
7050                         r = s->rule.ptr;
7051                         a = s->anchor.ptr;
7052                         log = s->log;
7053                 } else if (s == NULL) {
7054                         action = pf_test_rule(&r, &s, dir, kif,
7055                                               m, off, h, &pd, &a,
7056                                               &ruleset, NULL, inp);
7057                 }
7058                 break;
7059         }
7060
7061         default:
7062                 action = pf_test_state_other(&s, dir, kif, m, &pd);
7063                 if (action == PF_PASS) {
7064                         r = s->rule.ptr;
7065                         a = s->anchor.ptr;
7066                         log = s->log;
7067                 } else if (s == NULL) {
7068                         action = pf_test_rule(&r, &s, dir, kif, m, off, h,
7069                                               &pd, &a, &ruleset, NULL, inp);
7070                 }
7071                 break;
7072         }
7073
7074 done:
7075         if (n != m) {
7076                 m_freem(n);
7077                 n = NULL;
7078         }
7079
7080         /* handle dangerous IPv6 extension headers. */
7081         if (action == PF_PASS && rh_cnt &&
7082             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
7083                 action = PF_DROP;
7084                 REASON_SET(&reason, PFRES_IPOPTIONS);
7085                 log = 1;
7086                 DPFPRINTF(PF_DEBUG_MISC,
7087                     ("pf: dropping packet with dangerous v6 headers\n"));
7088         }
7089
7090         if ((s && s->tag) || r->rtableid)
7091                 pf_tag_packet(m, s ? s->tag : 0, r->rtableid);
7092
7093 #if 0
7094         if (dir == PF_IN && s && s->key[PF_SK_STACK])
7095                 m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK];
7096 #endif
7097
7098 #ifdef ALTQ
7099         /*
7100          * Generate a hash code and qid request for ALTQ.  A qid of 0
7101          * is allowed and will cause altq to select the default queue.
7102          */
7103         if (action == PF_PASS) {
7104                 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
7105                 if (pd.tos & IPTOS_LOWDELAY)
7106                         m->m_pkthdr.pf.qid = r->pqid;
7107                 else
7108                         m->m_pkthdr.pf.qid = r->qid;
7109                 m->m_pkthdr.pf.ecn_af = AF_INET6;
7110                 m->m_pkthdr.pf.hdr = h;
7111                 if (s) {
7112                         /* for fairq */
7113                         m->m_pkthdr.pf.state_hash = s->hash;
7114                         m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED;
7115                 }
7116         }
7117 #endif /* ALTQ */
7118
7119         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
7120             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
7121             (s->nat_rule.ptr->action == PF_RDR ||
7122             s->nat_rule.ptr->action == PF_BINAT) &&
7123             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
7124                 m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST;
7125
7126         if (dir == PF_IN && action == PF_PASS && r->divert.port) {
7127                 struct pf_divert *divert;
7128
7129                 if ((divert = pf_get_divert(m))) {
7130                         m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED;
7131                         divert->port = r->divert.port;
7132                         divert->addr.ipv6 = r->divert.addr.v6;
7133                 }
7134         }
7135
7136         if (log) {
7137                 struct pf_rule *lr;
7138
7139                 if (s != NULL && s->nat_rule.ptr != NULL &&
7140                     s->nat_rule.ptr->log & PF_LOG_ALL)
7141                         lr = s->nat_rule.ptr;
7142                 else
7143                         lr = r;
7144                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, reason, lr, a, ruleset,
7145                     &pd);
7146         }
7147
7148         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
7149         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
7150
7151         if (action == PF_PASS || r->action == PF_DROP) {
7152                 dirndx = (dir == PF_OUT);
7153                 r->packets[dirndx]++;
7154                 r->bytes[dirndx] += pd.tot_len;
7155                 if (a != NULL) {
7156                         a->packets[dirndx]++;
7157                         a->bytes[dirndx] += pd.tot_len;
7158                 }
7159                 if (s != NULL) {
7160                         if (s->nat_rule.ptr != NULL) {
7161                                 s->nat_rule.ptr->packets[dirndx]++;
7162                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
7163                         }
7164                         if (s->src_node != NULL) {
7165                                 s->src_node->packets[dirndx]++;
7166                                 s->src_node->bytes[dirndx] += pd.tot_len;
7167                         }
7168                         if (s->nat_src_node != NULL) {
7169                                 s->nat_src_node->packets[dirndx]++;
7170                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
7171                         }
7172                         dirndx = (dir == s->direction) ? 0 : 1;
7173                         s->packets[dirndx]++;
7174                         s->bytes[dirndx] += pd.tot_len;
7175                 }
7176                 tr = r;
7177                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
7178                 if (nr != NULL && r == &pf_default_rule)
7179                         tr = nr;
7180                 if (tr->src.addr.type == PF_ADDR_TABLE)
7181                         pfr_update_stats(tr->src.addr.p.tbl,
7182                             (s == NULL) ? pd.src :
7183                             &s->key[(s->direction == PF_IN)]->addr[0],
7184                             pd.af, pd.tot_len, dir == PF_OUT,
7185                             r->action == PF_PASS, tr->src.neg);
7186                 if (tr->dst.addr.type == PF_ADDR_TABLE)
7187                         pfr_update_stats(tr->dst.addr.p.tbl,
7188                             (s == NULL) ? pd.dst :
7189                             &s->key[(s->direction == PF_IN)]->addr[1],
7190                             pd.af, pd.tot_len, dir == PF_OUT,
7191                             r->action == PF_PASS, tr->dst.neg);
7192         }
7193
7194
7195         if (action == PF_SYNPROXY_DROP) {
7196                 m_freem(*m0);
7197                 *m0 = NULL;
7198                 action = PF_PASS;
7199         } else if (r->rt)
7200                 /* pf_route6 can free the mbuf causing *m0 to become NULL */
7201                 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
7202
7203         return (action);
7204 }
7205 #endif /* INET6 */
7206
7207 int
7208 pf_check_congestion(struct ifqueue *ifq)
7209 {
7210                 return (0);
7211 }